An extreme value search control method based on multi-scale time window adaptive injection

Abstract

The embodiment of the application provides a kind of based on multi-scale time window adaptive injection extreme value search control method, for the multi-time scale thermal dynamic characteristics (such as TEC fast response and crystal overall thermal balance slow response) existing in solid laser crystal temperature control system, by simultaneously injecting N frequency geometric progression distribution, fully separated sinusoidal disturbance signal in frequency domain, different scale thermal dynamic process is matched respectively;For the efficiency gradient estimate obtained by demodulating each scale, the signal-to-noise ratio based on real-time calculation is adaptively weighted and fused, so that the gradient information with high signal-to-noise ratio dominates in the fusion, so as to effectively suppress measurement noise;At the same time, according to the variance of the fused efficiency gradient, the frequency of each scale disturbance signal is dynamically scheduled, realizing multi-stage adaptive switching from coarse search to fine search to lock.

Description

Technical Field

[0001] This invention relates to the field of solid-state laser control and optimization technology, and in particular to an extremum search control method based on multi-scale time window adaptive injection. Background Technology

[0002] The output efficiency of solid-state lasers is significantly affected by the crystal's operating temperature, essentially due to the nonlinear modulation of parameters such as absorption cross-section, stimulated emission cross-section, and fluorescence lifetime caused by temperature. To automatically track the optimal operating temperature that has drifted due to factors such as pump power variations and crystal aging, Extremum Seeking Control (ESC), a real-time optimization method that does not require a system model, has been introduced into the field of laser temperature control. Its basic idea is to superimpose a sinusoidal perturbation signal of fixed frequency and amplitude at the operating point, estimate the gradient by demodulating the response of the system output (e.g., laser efficiency), and gradually optimize along the gradient direction. In a typical implementation, the temperature control system estimates the gradient based on the fixed perturbation frequency and updates the temperature setpoint accordingly to maximize efficiency.

[0003] Existing single-scale ESC methods face an insurmountable dilemma in perturbation frequency selection. Solid-state laser crystal thermal dynamics systems inherently possess multiple timescales: the response of the thermoelectric cooler (TEC) or water-cooling system is relatively fast (on the order of seconds), while the process of the crystal reaching thermal equilibrium is slow (on the order of minutes). When selecting a fixed perturbation frequency, if the frequency is too high, the slow thermal dynamic process of the system has not yet been established, leading to severe distortion in the demodulated efficiency gradient estimate and potentially incorrect optimization direction. If the frequency is too low, although it allows for waiting for thermal equilibrium to be established, the waiting time per cycle is too long, resulting in extremely slow convergence of the entire optimal temperature search process, failing to respond quickly to dynamic changes in the working environment. A single frequency cannot simultaneously accommodate the fast and slow dual-timescale characteristics of the system.

[0004] The gradient estimation accuracy of existing single-scale ESC methods is highly susceptible to measurement noise, especially exhibiting poor robustness at low frequencies. During the locking phase when the crystal temperature approaches its optimum, to minimize disturbances to the laser's normal operation, it is typically necessary to reduce the perturbation amplitude and use a lower-frequency signal. However, at this point, the efficiency gradient signal itself is already very weak, while the laser power meter measurement noise has a higher power spectral density at low frequencies, leading to a sharp drop in the signal-to-noise ratio. The gradient estimate is overwhelmed by noise, making it impossible for the system to accurately locate the optimal temperature point, significantly reducing steady-state control accuracy. Furthermore, the fixed optimization strategy of existing single-scale ESC methods cannot balance global search accuracy with local convergence speed. Their fixed perturbation parameters cannot automatically adjust the intensity of coarse and fine searches according to the real-time optimization process, resulting in either low global search efficiency or insufficient local convergence accuracy. This makes it difficult to achieve efficient and stable optimal value search on complex crystal temperature-efficiency nonlinear curves. Summary of the Invention

[0005] This invention provides an extreme value search control method based on multi-scale time window adaptive injection to solve the technical problems of existing single-scale extreme value search control methods in solid-state laser crystal temperature optimization, such as difficulty in selecting perturbation frequency, susceptibility of gradient estimation to measurement noise interference, and inability to simultaneously achieve fast convergence performance and steady-state robustness.

[0006] In a first aspect, embodiments of the present invention provide an extreme value search control method based on multi-scale time window adaptive injection, comprising: S1. Initialize the current optimal temperature estimate and initialize the temperature perturbation signal parameters at N scales, where N is an integer greater than or equal to 2. The temperature perturbation signal parameters include the frequency, amplitude, and phase values ​​corresponding to the N scales. The frequency values ​​corresponding to the N scales are distributed according to a geometric progression, and the ratio of the frequency values ​​of any adjacent scales is a constant. S2. Superimpose the temperature disturbance signals of N scales onto the current optimal temperature estimate to generate a composite temperature control command, and output the composite temperature control command to the crystal temperature control actuator. S3. Real-time acquisition of the laser's output efficiency, and based on the output efficiency, perform gradient demodulation and low-pass filtering on each scale to obtain the efficiency gradient estimate corresponding to each scale. S4. Calculate the signal-to-noise ratio (SNR) for each scale in real time, calculate the corresponding adaptive weights based on the SNR of each scale, and perform a weighted summation of the efficiency gradient estimates for each scale based on the adaptive weights to obtain the fusion efficiency gradient. S5. Update the current optimal temperature estimate based on the fusion efficiency gradient; S6. Real-time detection of the variance of the fusion efficiency gradient, and dynamic scheduling of the frequency values ​​corresponding to the temperature perturbation signals at each scale according to the magnitude of the variance, so as to switch the search phase.

[0007] Preferably, the ratio of the frequency values ​​of any adjacent scales is a first fixed value, and the first fixed value is less than or equal to 1 / 5; the amplitude values ​​decay geometrically, and the ratio of the amplitude values ​​of any adjacent scales is a second fixed value.

[0008] Preferably, the highest frequency value among the frequency values ​​corresponding to the N scales is determined based on the thermal response time constant of the crystal, and the perturbation period corresponding to the highest frequency value is less than or equal to the fast time constant.

[0009] Preferably, the real-time calculation of the signal-to-noise ratio for each scale includes: The noise power spectral density corresponding to each scale is estimated in real time using a sliding window variance estimator, and the signal-to-noise ratio is calculated based on the ratio of the noise power spectral density to the power of the efficiency gradient estimate corresponding to the scale.

[0010] Preferably, the search phase includes a coarse search phase, a fine search phase, and a locking phase; the search switching phase includes: When the variance is greater than the first preset threshold, switch to the coarse search stage; When the variance is less than or equal to the first preset threshold and greater than or equal to the second preset threshold, switch to the fine search stage; When the variance is less than the second preset threshold, the system switches to the locking phase.

[0011] Preferably, during the locking phase, the following is also included: The laser output efficiency is monitored in real time. When the change exceeds the preset efficiency mutation threshold, it is determined that the current optimal temperature estimate has deviated from the true optimal temperature point, and the process switches back to the coarse search stage.

[0012] Preferably, the frequency and amplitude values ​​of the coarse search stage are both greater than those of the fine search stage; the frequency value of the fine search stage is one-fifth of the frequency value of the coarse search stage, and the amplitude value of the fine search stage is half of the amplitude value of the coarse search stage.

[0013] Preferably, in step S5, updating the current optimal temperature estimate based on the fusion efficiency gradient includes adding the current optimal temperature estimate to the product of a preset step size gain and the fusion efficiency gradient.

[0014] Preferably, in step S3, the cutoff frequency of the low-pass filtering process is lower than the minimum frequency value corresponding to each scale.

[0015] Secondly, embodiments of the present invention provide an extreme value search control device based on multi-scale time window adaptive injection, comprising: The multi-scale perturbation generation module is used to initialize the current optimal temperature estimate and initialize temperature perturbation signal parameters at N scales, where N is an integer greater than or equal to 2. The temperature perturbation signal parameters include frequency values, amplitude values, and phase values ​​corresponding to the N scales. The frequency values ​​corresponding to the N scales are distributed according to a geometric progression, and the ratio of the frequency values ​​of any adjacent scales is a constant. The instruction synthesis module is used to superimpose temperature disturbance signals of N scales onto the current optimal temperature estimate to generate a composite temperature control instruction, and output the composite temperature control instruction to the crystal temperature control actuator. A multi-scale demodulation and filtering module is used to acquire the output efficiency of the laser in real time, and perform gradient demodulation and low-pass filtering on each scale based on the output efficiency to obtain the efficiency gradient estimate corresponding to each scale. The signal-to-noise ratio (SNR) estimation module is used to calculate the SNR corresponding to each scale in real time, calculate the corresponding adaptive weights based on the SNR of each scale, and perform a weighted summation of the efficiency gradient estimates of each scale based on the adaptive weights to obtain the fusion efficiency gradient. The temperature update module is used to update the current optimal temperature estimate based on the fusion efficiency gradient. The frequency scheduling module is used to detect the variance of the fusion efficiency gradient in real time, and dynamically schedule the frequency values ​​corresponding to the temperature perturbation signals at each scale according to the magnitude of the variance, so as to switch the search phase.

[0016] This invention provides an extreme value search control method and apparatus based on multi-scale time window adaptive injection. Addressing the multi-timescale thermodynamic characteristics (such as the fast TEC response and the slow overall thermal equilibrium response of the crystal) in solid-state laser crystal temperature control systems, this method simultaneously injects N sinusoidal perturbation signals with geometrically distributed frequencies and sufficient separation in the frequency domain to match the thermodynamic processes at different scales. The efficiency gradient estimates obtained from demodulation at each scale are adaptively weighted and fused based on the real-time calculated signal-to-noise ratio (SNR), ensuring that gradient information with a high SNR dominates the fusion, thereby effectively suppressing measurement noise. Simultaneously, the frequency of the perturbation signals at each scale is dynamically scheduled according to the variance of the fused efficiency gradient, achieving multi-stage adaptive switching from coarse search to fine search to locking. Compared with existing technologies, this method has the following advantages: (1) Multi-scale perturbation parallel injection overcomes the difficulty of frequency selection in traditional single-scale methods, ensuring both the real-time performance of fast dynamic response and the integrity of slow thermal equilibrium establishment, significantly improving the convergence speed to the optimal temperature point.

[0017] (2) The adaptive gradient fusion based on signal-to-noise ratio significantly reduces the interference of sensor noise such as power meter on gradient estimation, and improves the control accuracy and robustness in the steady-state locking stage.

[0018] (3) Through the frequency scheduling mechanism driven by gradient variance, the system can automatically adjust the search step size and disturbance characteristics according to the current optimization process. Without manual intervention, it can efficiently and stably track the optimal crystal temperature that drifts with the working conditions on the complex nonlinear efficiency curve, and finally achieve the continuous maximization of laser output efficiency. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The flowchart of the extreme value search control method based on multi-scale time window adaptive injection provided by the present invention is shown.

[0021] Figure 2 The flowchart for SNR adaptive weighted fusion execution provided in the embodiments of the present invention is shown.

[0022] Figure 3 The flowchart for adaptive injection perturbation execution is provided for an embodiment of the present invention.

[0023] Figure 4 The simulation comparison chart shows the convergence speed of the algorithm provided in this embodiment of the invention and the traditional single-scale ESC algorithm.

[0024] Figure 5 The simulation comparison chart shows the robustness of the algorithm provided in this embodiment of the invention compared with the traditional single-scale ESC algorithm.

[0025] Figure 6 This is a schematic diagram of the extreme value search control device based on multi-scale time window adaptive injection provided in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. 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.

[0027] Figure 1 This is a flowchart of an extreme value search control method based on multi-scale time window adaptive injection according to an embodiment of the present invention. The method of this embodiment is used to optimize the crystal operating temperature of a solid-state laser, with laser output efficiency as the objective function and crystal temperature setpoint as the control variable. Figure 1 The method includes: S1. Initialize the current optimal temperature estimate and initialize the temperature perturbation signal parameters at N scales, where N is an integer greater than or equal to 2. The temperature perturbation signal parameters include the frequency, amplitude, and phase values ​​corresponding to the N scales. The frequency values ​​corresponding to the N scales are distributed according to a geometric progression, and the ratio of the frequency values ​​of any adjacent scales is a constant.

[0028] Here, the current optimal temperature estimate is the initially set estimated optimal operating temperature of the crystal; N scales refer to the perturbation signal levels that adapt to the multi-timescale thermal dynamics of the crystal; the temperature perturbation signal parameters are the core parameters for controlling the perturbation signal; and the set of frequency values... Amplitude value set Phase value set These are the basic characteristic parameters of a sinusoidal disturbance signal. Each frequency satisfies The amplitude decreases geometrically. ,in, This is a frequency ratio parameter; the frequency is selected based on the crystal thermal response time constant. highest frequency satisfy This ensures that signals at each scale can effectively excite the crystal's thermal dynamic response. These parameters are used to generate the sinusoidal perturbation waveform injected into the subsequent temperature control system. The geometric progression distribution means that the frequency values ​​are arranged in a geometric series, and the ratio of adjacent scale frequency values ​​is a fixed value, meaning that the frequencies of each scale are set in a fixed proportion. In this embodiment, S1 establishes an initial optimal temperature estimation benchmark when the system starts up and presets N (N≥2) perturbation signal parameters at different time scales. To ensure effective separation of signals at each scale in the frequency domain, the frequency values ​​are distributed according to a geometric progression, that is, the ratio of adjacent scale frequencies is a fixed value (e.g., 1:5:25), so that high-frequency signals can quickly excite the crystal's fast-time dynamics (e.g., TEC response), and low-frequency signals can fully cover slow-time dynamics (e.g., overall crystal thermal equilibrium). This solves the problem in the prior art that single-scale perturbation frequencies cannot simultaneously adapt to the multi-time-scale characteristics of the system. Through multi-scale parameter initialization, it lays the foundation for subsequent parallel injection and demodulation. The technical effect is that the system has the ability to handle mixed fast and slow dynamics from the startup stage.

[0029] S2. Superimpose the temperature disturbance signals of N scales onto the current optimal temperature estimate to generate a composite temperature control command, and output the composite temperature control command to the crystal temperature control actuator.

[0030] Specifically, the temperature disturbance signal superposition involves arithmetically adding N scales of sinusoidal disturbance signals to the current optimal temperature estimate. The resulting composite temperature control command corresponds to the following formula:

[0031] in, θ ( t ) represents the current optimal crystal temperature estimate (unit: degrees Celsius). A i For the first i The magnitude of temperature disturbance on a scale (unit: degrees Celsius). ω i The disturbance frequency is expressed in rad / s. For the initial phase, this composite command T ( t The output is sent to the crystal temperature control actuator to drive the crystal temperature change.

[0032] S3. Real-time acquisition of the laser's output efficiency, and based on the output efficiency, perform gradient demodulation and low-pass filtering on each scale to obtain the efficiency gradient estimate corresponding to each scale.

[0033] Wherein, the laser output efficiency is the ratio of laser output power to pump input power, that is, the objective function is the laser output efficiency:

[0034] in, P out For laser output power, P in The pump input power is the objective function of the method in this embodiment of the invention. Gradient demodulation processing extracts the gradient information of efficiency versus temperature under corresponding scale perturbations from the real-time acquired efficiency signal. Low-pass filtering is used to filter out high-frequency noise in the signal to obtain the efficiency gradient estimate corresponding to each scale. The control objective of the method in this embodiment of the invention is to find the maximum point of the efficiency function, that is, to satisfy the formula... ,and Optimal temperature T ,in, The efficiency gradient varies with temperature. This is the second derivative of the efficiency function. This step acquires the laser output efficiency and performs gradient demodulation and low-pass filtering for each scale. This addresses the technical shortcomings of traditional single-scale extremum search control, which suffers from gradient estimation distortion and susceptibility to noise interference due to incomplete establishment of crystal thermal dynamics. This provides a reliable gradient estimation basis for subsequent gradient fusion and temperature updates.

[0035] S4. Calculate the signal-to-noise ratio (SNR) for each scale in real time, calculate the corresponding adaptive weights based on the SNR of each scale, and perform a weighted summation of the efficiency gradient estimates for each scale based on the adaptive weights to obtain the fusion efficiency gradient.

[0036] In this process, the signal-to-noise ratio (SNR) is the ratio of the signal power to the noise power of the efficiency gradient estimates at each scale. The adaptive weights are gradient fusion coefficients dynamically allocated based on the SNR at each scale. The weighted summation is the combined calculation of the efficiency gradient estimates at each scale according to the adaptive weights to obtain the fused efficiency gradient. The fused gradient information can more accurately reflect the overall trend of efficiency changes with temperature, adapting to the multi-timescale thermal dynamics of the crystal. Step S4 calculates the adaptive weights corresponding to each scale based on the real-time SNR and performs a weighted summation of the efficiency gradient estimates at each scale to obtain the fused efficiency gradient. This solves the technical defects of low SNR and insufficient gradient estimation accuracy in the low-frequency band of traditional single-scale extremum search control, effectively suppressing the interference of measurement noise on gradient estimation and improving the accuracy and robustness of gradient information.

[0037] S5. Update the current optimal temperature estimate based on the fusion efficiency gradient; the fusion efficiency gradient is the total gradient value after fusing multi-scale gradient information. Updating the current optimal temperature estimate is an operation that adjusts the estimated optimal temperature of the crystal based on the fusion efficiency gradient. The purpose of the update is to make the current temperature estimate gradually approach the desired temperature. Optimal temperature When the fusion gradient is positive, it indicates that the current temperature is lower than the optimal temperature, and the temperature estimate needs to be increased; when the fusion gradient is negative, it indicates that the current temperature is higher than the optimal temperature, and the temperature estimate needs to be decreased. Step S5 updates the current optimal temperature estimate based on the fusion efficiency gradient, addressing the technical shortcomings of traditional solid-state laser temperature control, which uses a fixed temperature setpoint and cannot track the optimal operating temperature that dynamically drifts with changes in pump power, crystal aging, and ambient temperature. This enables online dynamic adjustment of the crystal's optimal temperature, gradually approaching the true optimal temperature point, and completing the optimal temperature search without relying on a system model.

[0038] S6. Real-time detection of the variance of the fusion efficiency gradient, and dynamic scheduling of the frequency values ​​corresponding to the temperature perturbation signals at each scale according to the magnitude of the variance, so as to switch the search phase.

[0039] The variance of the fusion efficiency gradient is a statistical value representing the degree of fluctuation in the fusion gradient signal. The dynamic scheduling frequency value is the frequency value for adjusting the temperature perturbation signal at each scale based on the variance. ω i The operation involves switching between search phases, which is a process of adjusting the search strategy based on changes in gradient variance. When the gradient variance is large, it indicates that the current temperature estimate is far from the optimal temperature point, requiring a higher frequency perturbation signal to accelerate the search process. When the gradient variance is small, it indicates that the current temperature estimate is close to the optimal temperature point, requiring adjustment of the perturbation frequency to achieve a more accurate search. This step detects the variance of the fusion efficiency gradient in real time and dynamically schedules the frequency values ​​of temperature perturbation signals at each scale to switch search phases based on the variance magnitude. This addresses the technical shortcomings of traditional single-scale extreme value search control, which struggles to balance convergence speed and robustness and cannot adaptively match the crystal thermal equilibrium establishment process. It balances the speed and accuracy of optimal temperature search and adapts to the multi-timescale changes in crystal thermal dynamics.

[0040] Based on the above embodiments, as a preferred implementation, the ratio of the frequency values ​​of any adjacent scales is a first fixed value, and the first fixed value is less than or equal to 1 / 5; the amplitude values ​​decay geometrically, and the ratio of the amplitude values ​​of any adjacent scales is a second fixed value.

[0041] The first constant value refers to the frequency of perturbations at adjacent scales. ω i The first in i The ratio of scale perturbation frequencies, expressed in rad / s, and this ratio is not greater than 1 / 5; the geometrical attenuation of amplitude values ​​refers to the amplitude of adjacent scale perturbations. A i The frequency is attenuated at a fixed ratio, which is a second fixed value. In the parameter initialization step S1 of this embodiment, the frequencies of each scale are distributed at a fixed ratio not exceeding 1 / 5, and the amplitudes are attenuated at a fixed ratio. The frequency ratio constraint ensures that the perturbation signals of each scale are fully separated in the frequency domain, avoiding demodulation interference caused by the overlap of the spectrum of signals of different scales. At the same time, the amplitude attenuation makes the high-frequency scale use a large amplitude perturbation and the low-frequency scale use a small amplitude perturbation, which is adapted to the multi-timescale thermal dynamic characteristics of the crystal's fast response temperature control and slow thermal balance. This solves the technical defects of existing multi-scale schemes, such as gradient demodulation distortion caused by insufficient frequency spacing and poor crystal temperature stability caused by large low-frequency perturbations. This parameter setting can ensure that the gradient demodulation processing of each scale in the subsequent step S3 does not interfere with each other, improves the accuracy of efficiency gradient estimation, and reduces the impact of large fluctuations in crystal temperature caused by low-frequency perturbations. It balances gradient estimation accuracy and temperature control stability, and provides a reliable foundation for subsequent fusion gradient calculation and temperature update.

[0042] Based on the above embodiments, as a preferred implementation, the highest frequency value among the frequency values ​​corresponding to the N scales is determined according to the thermal response time constant of the crystal, and the perturbation period corresponding to the highest frequency value is less than or equal to the fast time constant.

[0043] Among them, the frequency values ​​corresponding to the N scales are the frequencies of each level of the multi-scale perturbation signal, with the highest frequency value being... ω max For all ω i The maximum value in the range corresponds to a disturbance period of . T max =2π / ω max (i.e., the duration of a sinusoidal perturbation), the thermal response time constant of a crystal specifically refers to the fast response time constant of the temperature control actuator (such as a TEC semiconductor cooler or water cooling system) in a solid-state laser temperature control system. τ fast (In seconds) represents the fastest thermal dynamic response time in the system. In this embodiment, during parameter initialization in step S1, the highest frequency value is determined based on the fast time constant of the crystal temperature control system, ensuring that the disturbance period corresponding to the highest frequency is less than or equal to this fast time constant. T max ≤ τ fast This invention ensures that high-frequency disturbance signals can be effectively responded to by the temperature control actuator. It addresses the problem caused by unreasonable setting of the highest frequency in existing multi-scale extremum search schemes: if the highest frequency is too low, its disturbance period is greater than the fast time constant, causing the temperature control actuator to be unable to effectively respond to high-frequency disturbances, and the crystal temperature to fail to follow the changes in high-frequency disturbances. Consequently, the gradient demodulation processing at the fast time scale in step S3 cannot extract the true efficiency gradient information, resulting in distorted gradient estimation. By limiting the disturbance period of the highest frequency to within the fast time constant, the scheme of this embodiment can accurately excite the thermal dynamics at the fast time scale, providing true temperature-efficiency change information for the multi-scale gradient demodulation in step S3, improving the accuracy of fast-scale gradient estimation, while not affecting the excitation of the slow thermal equilibrium dynamics of the crystal at the low-frequency scale. This balances the excitation effect of multi-scale thermal dynamics, providing a reliable foundation for subsequent gradient fusion and temperature updates, and improving the robustness and temperature control accuracy of the entire method.

[0044] Based on the above embodiments, as a preferred implementation method, such as Figure 2 As shown, the real-time calculation of the signal-to-noise ratio corresponding to each scale includes: The noise power spectral density corresponding to each scale is estimated in real time using a sliding window variance estimator, and the signal-to-noise ratio is calculated based on the ratio of the noise power spectral density to the power of the efficiency gradient estimate corresponding to the scale.

[0045] The sliding window variance estimator is a computational unit that uses a fixed-length sliding window to statistically analyze signal fluctuations and quantify noise characteristics in real time. The noise power spectral density is the power distribution characteristic value of the power meter measurement noise in the perturbation signal at each scale. The power of the efficiency gradient estimator is the power value of the effective efficiency gradient signal extracted at the corresponding scale. The signal-to-noise ratio (SNR) is the ratio of the effective gradient signal power to the noise power spectral density. The corresponding SNR calculation formula is as follows:

[0046] in, SNR i For the first i Scale signal-to-noise ratio, η i ( t ) is the first i Efficiency gradient estimator corresponding to scale The signal power of the gradient estimator, The real-time estimation of the variance by the sliding window variance estimator i The method in this embodiment of the invention obtains the noise power spectral density at each scale in real time through a sliding window variance estimator. Then, the power of the efficiency gradient estimate. The signal-to-noise ratio is calculated as the ratio of the signal to the noise power spectral density. SNR i This invention addresses the technical shortcomings of traditional single-scale extremum search control, which cannot accurately estimate real-time noise and suffers from large deviations in signal-to-noise ratio calculation, leading to interference from power meter measurement noise in gradient estimation. It can obtain the signal-to-noise ratio values ​​of each scale in real time and accurately provide a precise basis for subsequent adaptive weight allocation, ensuring that scales with high signal-to-noise ratios have a greater weight in gradient fusion. It effectively suppresses the interference of power meter measurement noise on temperature optimization and improves the accuracy and robustness of gradient calculation in fusion efficiency.

[0047] Subsequently, adaptive weights are calculated based on the signal-to-noise ratio at each scale. The adaptive weight fusion formula is as follows:

[0048] in, w i ( t ) is the first i The scale is adaptively weighted, where N is the total number of perturbation scales. Finally, the efficiency gradient estimates for each scale are weighted and summed according to the adaptive weights to obtain the fusion efficiency gradient, which is given by the formula:

[0049] in, This invention addresses the technical shortcomings of traditional single-scale extremum search control, such as the inability to accurately estimate noise in real time, large deviations in signal-to-noise ratio calculation leading to unreasonable weight allocation, and susceptibility of gradient estimation to power meter measurement noise interference. It improves the stability of noise estimation through a fixed-length sliding window and ensures that scales with high signal-to-noise ratio and good signal quality receive greater weights through an adaptive weight allocation mechanism based on signal-to-noise ratio. This effectively suppresses the interference of measurement noise on gradient estimation, improves the accuracy and robustness of the fusion efficiency gradient, and provides reliable and clean gradient signal support for updating the optimal temperature estimate in subsequent step S5, thus ensuring the accuracy and stability of crystal temperature optimization control.

[0050] The fusion efficiency gradient is obtained by weighted summation of the efficiency gradient estimates at each scale:

[0051] The noise suppression effect of multi-scale fusion follows the formula:

[0052] That is, the variance of the total gradient estimate after fusion is strictly less than the noise variance of any single scale. The stability of noise estimation is improved by using a fixed-length sliding window. The adaptive weight allocation mechanism based on signal-to-noise ratio prioritizes the retention of gradient information of high signal-to-noise ratio scales. At the same time, the noise suppression characteristics of multi-scale fusion theoretically ensure that the noise level of the gradient signal after fusion is lower than that of all single-scale schemes, effectively suppressing the interference of measurement noise on gradient estimation. This provides low-noise, high-reliability gradient signal support for the temperature update in the subsequent step S5, significantly improving the accuracy and robustness of crystal temperature optimization control. It is especially suitable for actual laser systems where the power meter has measurement noise.

[0053] Based on the above embodiments, as a preferred implementation, the search phase includes a coarse search phase, a fine search phase, and a locking phase; the switching search phase includes: When the variance is greater than the first preset threshold, switch to the coarse search stage.

[0054] When the variance is less than or equal to the first preset threshold and greater than or equal to the second preset threshold, switch to the fine search stage.

[0055] When the variance is less than the second preset threshold, the system switches to the locking phase.

[0056] Specifically, the variance of the fusion efficiency gradient is the real-time fluctuation statistical value of the fusion efficiency gradient obtained through multi-scale weighted fusion in step S4, reflecting the degree of deviation between the current crystal temperature estimate and the true optimal temperature point. The first preset threshold and the second preset threshold are two pre-set variance thresholds. The variance is divided into three intervals as the judgment condition for stage switching. The coarse search stage, the fine search stage, and the locking stage are three control stages that are advanced in sequence, corresponding to the control objectives of large-scale rapid positioning, high-precision approximation, and stable maintenance of the optimal temperature, respectively. This scheme achieves automatic stage switching by detecting the variance of the fusion efficiency gradient in real time and comparing it with the two preset thresholds: when the variance is greater than the first preset threshold, it is determined that the current temperature estimate is far from the optimal temperature point, and the process switches to the coarse search stage; when the variance is less than or equal to the first preset threshold and greater than or equal to the second preset threshold, the process switches to the second preset threshold. The current temperature estimate is close to the optimal temperature point, so the system switches to the fine search stage. When the variance is less than the second preset threshold, the current temperature estimate is determined to be stable near the optimal temperature, and the system switches to the locking stage. This solution solves the technical defects of traditional single-scale extreme value search control, which has a fixed disturbance frequency and cannot adaptively adjust the control strategy according to the search process, making it difficult to balance convergence speed, control accuracy, and stability. It achieves three-stage automatic switching through real-time feedback of gradient variance. The coarse search stage can quickly narrow the optimal temperature range and improve the initial search efficiency. The fine search stage can achieve accurate approximation of the optimal temperature and improve control accuracy. The locking stage can maintain temperature stability and avoid unnecessary disturbances from affecting the stability of laser output. The entire process does not require manual intervention and can automatically adjust the control strategy according to the crystal thermal dynamics and operating conditions, effectively improving the robustness and temperature control effect of extreme value search control.

[0057] Based on the above embodiments, as a preferred implementation, when in the locking phase, it further includes: The laser output efficiency is monitored in real time. When the change exceeds the preset efficiency mutation threshold, it is determined that the current optimal temperature estimate has deviated from the true optimal temperature point, and the process switches back to the coarse search stage.

[0058] The locking phase is the control phase when the variance of the fusion efficiency gradient is less than a second preset threshold. At this time, the multi-scale temperature perturbation frequency approaches zero to maintain the crystal temperature stable near the current optimal temperature estimate. The change in laser output efficiency refers to the deviation between the real-time acquired laser output efficiency and the historical stable efficiency benchmark value, used to reflect the degree of efficiency fluctuation. The efficiency mutation threshold is a preset critical value, determined based on the efficiency fluctuation range when the laser is working normally, used to distinguish between normal efficiency fluctuations and abnormal changes caused by the drift of the optimal temperature point. The current optimal temperature estimate is the estimated value of the optimal operating temperature of the crystal obtained through iterative updates during the control process. The true optimal temperature point is the actual temperature point at which the laser output efficiency reaches its maximum value under the current operating conditions of the crystal, and it will dynamically change with factors such as pump power, crystal aging, and ambient temperature. Switching back to the coarse search phase means adjusting the control mode from the locking phase to the coarse search phase, and using high-frequency, large-amplitude multi-scale perturbation signals to quickly locate the new optimal temperature range. In this embodiment, the locking phase... The system continuously monitors changes in laser output efficiency. When the change exceeds a preset efficiency mutation threshold, it determines that the current optimal temperature estimate has deviated from the true optimal temperature point. At this point, it automatically switches back to the coarse search stage and restarts the large-scale optimal temperature search process. In existing technologies, extreme value search control only maintains a fixed temperature setpoint during the locking stage, which cannot cope with the drift of the true optimal temperature point caused by changes in operating conditions. When pump power, crystal aging, or ambient temperature changes, the original optimal temperature estimate becomes invalid, and the control algorithm cannot recognize this change, continuing to maintain the deviated temperature, resulting in a decrease in laser output efficiency and an inability to recover to the optimal operating state. This solution actively identifies the drift of the optimal temperature by monitoring efficiency changes in real time. When abnormal changes are detected, it automatically triggers a re-search, adaptively tracking changes in operating conditions without manual intervention. This ensures that the laser always operates near the true optimal temperature, maintaining stable high-efficiency output, improving the robustness and long-term operational reliability of the control algorithm, and avoiding efficiency loss caused by changes in operating conditions.

[0059] Based on the above embodiments, as a preferred implementation, the frequency and amplitude values ​​of the coarse search stage are both greater than those of the fine search stage; the frequency value of the fine search stage is one-fifth of the frequency value of the coarse search stage, and the amplitude value of the fine search stage is half of the amplitude value of the coarse search stage.

[0060] In existing technologies, the stage switching of extreme value search control lacks a clear quantitative relationship of perturbation parameters. When the perturbation frequency or amplitude is insufficient in the coarse search stage, it cannot effectively excite the multi-timescale thermal dynamics of the crystal, resulting in insignificant efficiency gradient changes and difficulty in quickly locating the optimal temperature range. When the perturbation frequency or amplitude is too large in the fine search stage, it will induce temperature oscillations near the optimal temperature, reducing control accuracy and laser output stability. At the same time, the lack of a frequency and amplitude matching ratio makes it impossible to take into account the control objectives of different stages. The embodiments of the present invention clarify the frequency and amplitude ratio relationship between the coarse and fine search stages. The high-frequency large-amplitude perturbation in the coarse search stage can quickly excite the full-timescale thermal dynamics of the crystal. To improve initial search efficiency, low-frequency small-amplitude perturbations during the fine search phase can achieve refined gradient estimation near the optimal temperature. This ensures both the signal-to-noise ratio of the gradient signal and reduces the impact of perturbations on crystal temperature stability. Meanwhile, the one-fifth frequency ratio ensures sufficient separation of multi-scale signals in the frequency domain, avoiding spectral overlap between signals of different scales and guaranteeing the accuracy of subsequent gradient demodulation processing. The halved amplitude setting adapts to the convergence characteristics of crystal thermal dynamics as the search progresses, reducing unnecessary temperature fluctuations and providing conditions for a smooth transition in the subsequent locking phase. This effectively balances the convergence speed, control accuracy, and temperature control stability of extreme value search control, thereby improving overall control performance.

[0061] Based on the above embodiments, as a preferred implementation, in step S5, updating the current optimal temperature estimate according to the fusion efficiency gradient includes: adding the current optimal temperature estimate to the product of a preset step size gain and the fusion efficiency gradient.

[0062] The fusion efficiency gradient is the total gradient value obtained by multi-scale adaptive weighted fusion in step S4. η total ( t The current optimal temperature estimate is the predicted value of the optimal operating temperature of the crystal, which is updated iteratively. θ ( t Step gain The adjustment coefficient for temperature updates, control period Δ t The time interval for temperature updates is defined by the step size stability constraint, which is a step size value condition set based on the thermal response characteristics of the crystal. Lyapunov stability analysis is a theoretical method for verifying the stability of the system, and the candidate function is the energy function selected during the analysis. V ( e , θ Crystal temperature tracking error is the deviation between the actual temperature and the set temperature value; temperature optimization error is the deviation between the set temperature value and the true optimal temperature point; exponential convergence is the characteristic of error decaying exponentially with time; upper bound of gradient estimation error is the maximum value of error during gradient estimation; and bounded neighborhood is the finite range within which the error eventually converges.

[0063] In step S5 of this embodiment, the current optimal temperature estimate is updated according to the fusion efficiency gradient, and the crystal temperature setpoint update formula is: ,in For step size gain, To control the cycle. Due to the time lag in the crystal's thermal response, the step size... The selection must meet the following requirements. To ensure stability, among which For the curvature of the efficiency curve, This represents the equivalent disturbance amplitude.

[0064] Lyapunov stability analysis of the system shows that selecting candidate functions... ,in For crystal temperature tracking error, under appropriate parameter conditions, the temperature optimization error satisfies That is, the exponential convergence to the state of... Within a bounded neighborhood of radius , where This is the upper bound of the gradient estimation error.

[0065] Existing traditional solid-state laser temperature control uses a fixed temperature setpoint, which cannot dynamically adjust the optimal temperature according to changes in operating conditions. The temperature update method of traditional single-scale extreme value search control is crude and prone to over-update or slow convergence, and cannot adapt to the multi-timescale thermal dynamic characteristics of crystal fast-slow coupling. The embodiment of this invention uses linear iterative update of the optimal temperature estimate driven by the fusion efficiency gradient, which allows the temperature estimate to smoothly approach the true optimal temperature point along the efficiency increase direction, without relying on the crystal thermal dynamic system model. At the same time, the reasonable selection of step size gain can avoid oscillation or divergence in temperature update, taking into account the convergence speed of optimal temperature search and control stability, realizing real-time tracking of the dynamically drifting optimal temperature point, and continuously ensuring that the laser output efficiency is maintained at the optimal level.

[0066] Based on the above embodiments, as a preferred implementation, in step S3, the cutoff frequency of the low-pass filtering process is lower than the minimum frequency value corresponding to each scale.

[0067] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Specific implementation scenario: Crystal temperature optimization control of a solid-state laser based on CLBO crystal frequency doubling. System parameters: The wavelength of the input crystal is 532nm, and the power of the laser is... Crystal temperature control range 130°C~150°C, TEC temperature control system with fast time constant. The slow time constant of the overall thermal equilibrium of the crystal The power meter has a sampling rate of 10Hz and a measurement noise standard deviation of approximately ±3%.

[0068] System initialization, such asFigure 3 As shown, based on the crystal thermal response time constant Set temperature perturbation parameters at three scales: Scale 1 (coarse scale, fast response): , (Period 10s, approximately) ); Scale 2 (Mesoscale): , (Period 50s); Scale 3 (Fine scale, slow response): , (Period 250s, approximately) Frequency ratio ,satisfy Constraints. Initialize temperature setpoint. (Nominal operating temperature), clear the gradient buffer to zero, and set phase=1 (coarse search).

[0069] Multi-scale temperature perturbation injection superimposes sinusoidal perturbations of three scales onto the crystal temperature setpoint: (Unit: degrees Celsius). The composite temperature command is sent to the TEC controller, and the TEC drives the crystal temperature to track this set value. Laser output power is simultaneously acquired. and laser input power Calculate real-time efficiency .

[0070] Scalar efficiency gradient estimation, with demodulation performed separately for each scale: Then, after passing through a cutoff frequency of The efficiency gradient estimate is obtained by using a low-pass filter. The noise variance at each scale is estimated in real time using a sliding window with a length of 5 perturbation periods. This is to distinguish between real gradient signals and measurement noise.

[0071] Adaptive weight fusion, calculating at each scale Normalization yields the weights Fusion efficiency gradient In the initial stage of the coarse search, scale 1 (high frequency) dominates due to its fast response and high signal-to-noise ratio, quickly locating the approximate range of the optimal temperature. As the search progresses, scale 3 (low frequency) gradually becomes dominant, accurately locating the optimal temperature. .

[0072] Frequency scheduling and phase switching, real-time calculation of efficiency gradient variance .when At this point, the fine-search phase begins, reducing all scale frequencies to 1 / 5 of their original values ​​and halving the amplitude to minimize disturbances to the normal operation of the laser. At this point, the system enters the lockout phase, the perturbation frequency approaches zero, and the optimal crystal temperature setpoint is maintained. .

[0073] Convergence and locking: During the locking phase, the system continuously monitors the change in laser output efficiency. When pump power changes, crystal aging, or ambient temperature fluctuations cause... If the optimal temperature point is determined to have drifted, the system will automatically return to the coarse search stage to re-search for a new optimal crystal temperature. This enables continuous adaptive tracking of the optimal operating temperature. Figure 4 This is a comparison of the convergence speed simulation results between the algorithm of this invention and the traditional single-scale ESC algorithm; Figure 5 A comparison of the robustness simulation results of the algorithm in this embodiment of the invention with the traditional single-scale ESC algorithm.

[0074] Secondly, embodiments of the present invention provide an extremum search control device based on multi-scale time window adaptive injection, such as... Figure 6 As shown, the device 600 includes: The multi-scale perturbation generation module 610 is used to initialize the current optimal temperature estimate and initialize temperature perturbation signal parameters for N scales, where N is an integer greater than or equal to 2. The temperature perturbation signal parameters include frequency values, amplitude values, and phase values ​​corresponding to the N scales. The frequency values ​​corresponding to the N scales are distributed according to a geometric progression, and the ratio of the frequency values ​​of any adjacent scales is a constant.

[0075] The instruction synthesis module 620 is used to superimpose temperature disturbance signals of N scales onto the current optimal temperature estimate to generate a composite temperature control instruction, and output the composite temperature control instruction to the crystal temperature control actuator.

[0076] The multi-scale demodulation and filtering module 630 is used to acquire the output efficiency of the laser in real time, and perform gradient demodulation and low-pass filtering on each scale based on the output efficiency to obtain the efficiency gradient estimate corresponding to each scale.

[0077] The signal-to-noise ratio estimation module 640 is used to calculate the signal-to-noise ratio corresponding to each scale in real time, calculate the corresponding adaptive weights based on the signal-to-noise ratios of each scale, and perform a weighted summation of the efficiency gradient estimates of each scale based on the adaptive weights to obtain the fusion efficiency gradient.

[0078] Temperature update module 650 is used to update the current optimal temperature estimate based on the fusion efficiency gradient.

[0079] The frequency scheduling module 660 is used to detect the variance of the fusion efficiency gradient in real time, and dynamically schedule the frequency values ​​corresponding to the temperature disturbance signals at each scale according to the magnitude of the variance, so as to switch the search phase.

[0080] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An extreme value search control method based on multi-scale time window adaptive injection, characterized in that, include: S1. Initialize the current optimal temperature estimate and initialize the temperature perturbation signal parameters at N scales, where N is an integer greater than or equal to 2. The temperature perturbation signal parameters include the frequency, amplitude, and phase values ​​corresponding to the N scales. The frequency values ​​corresponding to the N scales are distributed according to a geometric progression, and the ratio of the frequency values ​​of any adjacent scales is a constant. S2. Superimpose the temperature disturbance signals of N scales onto the current optimal temperature estimate to generate a composite temperature control command, and output the composite temperature control command to the crystal temperature control actuator. S3. Real-time acquisition of the laser's output efficiency, and based on the output efficiency, perform gradient demodulation and low-pass filtering on each scale to obtain the efficiency gradient estimate corresponding to each scale. S4. Calculate the signal-to-noise ratio (SNR) for each scale in real time, calculate the corresponding adaptive weights based on the SNR of each scale, and perform a weighted summation of the efficiency gradient estimates for each scale based on the adaptive weights to obtain the fusion efficiency gradient. S5. Update the current optimal temperature estimate based on the fusion efficiency gradient; S6. Real-time detection of the variance of the fusion efficiency gradient, and dynamic scheduling of the frequency values ​​corresponding to the temperature perturbation signals at each scale according to the magnitude of the variance, so as to switch the search phase.

2. The extreme value search control method based on multi-scale time window adaptive injection according to claim 1, characterized in that, The ratio of the frequency values ​​of any adjacent scale is a first fixed value, and the first fixed value is less than or equal to 1 / 5; the amplitude value decays geometrically, and the ratio of the amplitude values ​​of any adjacent scale is a second fixed value.

3. The extreme value search control method based on multi-scale time window adaptive injection according to claim 1, characterized in that, The highest frequency value among the frequency values ​​corresponding to the N scales is determined based on the thermal response time constant of the crystal, and the perturbation period corresponding to the highest frequency value is less than or equal to the fast time constant.

4. The extreme value search control method based on multi-scale time window adaptive injection according to claim 1, characterized in that, The real-time calculation of the signal-to-noise ratio for each scale includes: The noise power spectral density corresponding to each scale is estimated in real time using a sliding window variance estimator, and the signal-to-noise ratio is calculated based on the ratio of the noise power spectral density to the power of the efficiency gradient estimate corresponding to the scale.

5. The extreme value search control method based on multi-scale time window adaptive injection according to claim 1, characterized in that, The search phase includes a coarse search phase, a fine search phase, and a locking phase; the search switching phase includes: When the variance is greater than the first preset threshold, switch to the coarse search stage; When the variance is less than or equal to the first preset threshold and greater than or equal to the second preset threshold, switch to the fine search stage; When the variance is less than the second preset threshold, the system switches to the locking phase.

6. The extreme value search control method based on multi-scale time window adaptive injection according to claim 5, characterized in that, When in the locked phase, it also includes: The laser output efficiency is monitored in real time. When the change exceeds the preset efficiency mutation threshold, it is determined that the current optimal temperature estimate has deviated from the true optimal temperature point, and the process switches back to the coarse search stage.

7. The extreme value search control method based on multi-scale time window adaptive injection according to claim 5, characterized in that, The frequency and amplitude values ​​of the coarse search stage are both greater than those of the fine search stage; the frequency value of the fine search stage is one-fifth of the frequency value of the coarse search stage, and the amplitude value of the fine search stage is half of the amplitude value of the coarse search stage.

8. The extreme value search control method based on multi-scale time window adaptive injection according to claim 1, characterized in that, In step S5, updating the current optimal temperature estimate based on the fusion efficiency gradient includes adding the current optimal temperature estimate to the product of a preset step size gain and the fusion efficiency gradient.

9. The extreme value search control method based on multi-scale time window adaptive injection according to claim 1, characterized in that, In step S3, the cutoff frequency of the low-pass filtering process is lower than the minimum frequency value corresponding to each scale.

Images