Control method and system of all-solid-state high repetition frequency Q-switched laser

Abstract

The invention relates to the technical field of laser regulation and control, in particular to a control method and system of an all-solid-state high repetition frequency Q-switched laser. According to the control method of the all-solid-state high repetition frequency Q-switched laser, the all-solid-state high repetition frequency Q-switched laser comprises a TEC module, a Q-switched module and a wavelength adjusting module, and the control method of the all-solid-state high repetition frequency Q-switched laser comprises the steps that a target model is constructed; the target model is constructed based on historical operation parameters of the all-solid-state high-repetition-frequency Q-switched laser to obtain current operation parameters of the all-solid-state high-repetition-frequency Q-switched laser; preprocessing the current operation parameters; based on the target model and the current operation parameters, determining a first working instruction corresponding to the TEC module, a second working instruction corresponding to the Q-switching module and a third working instruction corresponding to the wavelength adjusting module; the TEC module is controlled to work according to the first working instruction, the Q-switching module is controlled to work according to the second working instruction, and the wavelength adjusting module is controlled to work according to the third working instruction.

Description

Technical Field

[0001] The present application relates to the technical field of laser control, and in particular to a control method and system for an all-solid-state high-repetition-rate Q-switched laser. Background Art

[0002] At present, in some high-precision fields, there are great requirements for the output accuracy and stability of lasers, such as medical treatment, communication, material processing and environmental monitoring. Therefore, it is necessary to improve the stability of the laser output.

[0003] In the related technology, in order to achieve controllable laser output and improve the stability of laser output, a wavelength tunable all-solid-state high repetition rate Q-switched laser is used. Q-switched laser can achieve high repetition rate pulse output by controlling pulse width and frequency to meet the needs of high-speed and precision processing.

[0004] In the process of implementing the relevant technology, at least the following problems are found in the relevant technology: The wavelength tunable all-solid-state high repetition rate Q-switched laser in the related art requires manual adjustment of the laser, which has poor precision and is prone to inaccurate adjustment, resulting in the output wavelength of the laser being inaccurate and having poor stability. Summary of the invention

[0005] The embodiments of the present application disclose a control method and system for an all-solid-state high repetition rate Q-switched laser, which maintain the stability and consistency of the output laser of the all-solid-state high repetition rate Q-switched laser, improve the quality of the output laser, and improve the accuracy of the wavelength of the output laser.

[0006] In some embodiments, a control method for an all-solid-state high repetition rate Q-switched laser is provided. The all-solid-state high repetition rate Q-switched laser includes a TEC module, a Q-switching module and a wavelength adjustment module. The control method for the all-solid-state high repetition rate Q-switched laser includes: constructing a target model; the target model is constructed based on historical operating parameters of the all-solid-state high repetition rate Q-switched laser; obtaining current operating parameters of the all-solid-state high repetition rate Q-switched laser; preprocessing the current operating parameters; based on the target model and the current operating parameters, determining a first working instruction corresponding to the TEC module, a second working instruction corresponding to the Q-switching module, and a third working instruction corresponding to the wavelength adjustment module; controlling the TEC module to work according to the first working instruction, controlling the Q-switching module to work according to the second working instruction, and controlling the wavelength adjustment module to work according to the third working instruction.

[0007] The control method of the all-solid-state high repetition rate Q-switched laser provided in the embodiment of the present application can adjust the TEC module, Q-switching module and wavelength adjustment module of the all-solid-state high repetition rate Q-switched laser based on the current operating parameters of the all-solid-state high repetition rate Q-switched laser, thereby adjusting the beam quality, wavelength and laser pulse of the laser output by the all-solid-state high repetition rate Q-switched laser, realizing intelligent wavelength and pulse control, efficient dynamic adjustment and optimization, and enhancing the stability and adaptability of the output laser. Maintain the stability and consistency of the laser output by the all-solid-state high repetition rate Q-switched laser, improve the quality of the output laser, and improve the accuracy of the wavelength of the output laser.

[0008] Optionally, the wavelength adjustment module includes a first birefringent filter and a second birefringent filter, and constructs a target model, including: obtaining multiple groups of historical operating parameters of the all-solid-state high-repetition-rate Q-switched laser; the historical operating parameters include historical working parameters and historical output parameters corresponding to the historical working parameters; the historical working parameters include historical pump source current, historical crystal temperature, historical Q-switching parameters of the Q-switching module, and historical optical axis angles of the first birefringent filter and the second birefringent filter in the wavelength adjustment module; the historical output parameters include historical output spectrum, historical output pulse width, historical output repetition frequency, and historical output power; preprocessing each group of historical operating parameters; generating a proxy identifier based on active feedback learning technology and historical operating parameters; adding the proxy identifier and multiple groups of historical operating parameters to a training data set; and using the training data set to train a self-adjusting convolutional neural network to determine the target model.

[0009] Optionally, preprocessing each set of historical operating parameters includes: standardizing historical output parameters in the historical operating parameters based on a preset formula; and / or filtering the spectrum in the historical output parameters.

[0010] Optionally, standardizing the historical output parameters in the historical operating parameters includes: sequentially taking the historical output spectrum, historical output pulse width, historical output repetition frequency, and historical output power in the historical output parameters as the target historical parameters ; Calculate the average value of the target historical parameters among multiple sets of historical operating parameters and standard deviation ; The preset formulas include: ; in, is the standardized historical output parameter.

[0011] Optionally, based on active feedback learning technology and historical operating parameters, an agent identifier is generated, including: based on difference-driven learning technology, constructing multiple groups of positively correlated sample groups and negatively correlated sample groups of historical operating parameters; the agent identifier includes the positively correlated sample group and the negatively correlated sample group; the positively correlated sample group and the negatively correlated sample group are both obtained by transforming the historical operating parameters.

[0012] Optionally, the loss function for difference-driven learning is: ; in, is the agent identifier, For sample The feature mapping function is For sample The feature map of For sample The feature map of is the Euclidean distance between positively correlated sample groups, is the Euclidean distance between negatively correlated sample groups, is the preset boundary parameter, is the total number of positively correlated sample groups and negatively correlated sample groups.

[0013] Optionally, a self-adjusting convolutional neural network is trained using a training data set to determine a target model, including: constructing a self-adjusting convolutional neural network; the self-adjusting convolutional neural network includes a convolutional layer, a pooling layer, and a fully connected layer; inputting the training data set into the self-adjusting convolutional neural network; the convolutional layer extracts feature maps in the data based on a multi-level convolution operation; the pooling layer converts a first size of the feature map into a second size based on a downsampling operation, and the second size is smaller than the first size; extracting feature information corresponding to the feature map; the fully connected layer realizes the final classification and regression of the feature information based on a linear transformation.

[0014] Optionally, after the fully connected layer realizes the final classification and regression of feature information based on linear transformation, it also includes: optimizing the self-adjusting convolutional neural network based on a dynamic learning rate optimization algorithm or a performance optimization algorithm.

[0015] Optionally, the update formula of the dynamic learning rate optimization algorithm is: ; in, are the model parameters of the self-tuning convolutional neural network, is the learning rate, is the momentum estimate of the gradient, is the second moment estimate of the gradient, is a constant to prevent division by zero.

[0016] In some embodiments, a control system for an all-solid-state high repetition rate Q-switched laser is provided, comprising: an all-solid-state high repetition rate Q-switched laser, configured to output current operating parameters, the all-solid-state high repetition rate Q-switched laser comprising a TEC module, a Q-switching module and a wavelength adjustment module; a control module, communicatively connected to the all-solid-state high repetition rate Q-switched laser, the control module being configured to construct a target model; the target model being constructed based on historical operating parameters of the all-solid-state high repetition rate Q-switched laser; obtaining current operating parameters of the all-solid-state high repetition rate Q-switched laser; preprocessing the current operating parameters; inputting the current operating parameters into the target model, the target model outputting a first working instruction corresponding to the TEC module, a second working instruction corresponding to the Q-switching module and a third working instruction corresponding to the wavelength adjustment module; controlling the TEC module to work according to the first working instruction, controlling the Q-switching module to work according to the second working instruction, and controlling the wavelength adjustment module to work according to the third working instruction.

[0017] It can be understood that the beneficial effects that can be achieved by the technical solution provided by the control system of the all-solid-state high repetition rate Q-switched laser provided above can be referred to the beneficial effects in the control method of the all-solid-state high repetition rate Q-switched laser and any of its optional implementation methods, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 A structural block diagram of a control system for an all-solid-state high repetition rate Q-switched laser provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of an all-solid-state high repetition rate Q-switched laser provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a control system for an all-solid-state high repetition rate Q-switched laser provided in an embodiment of the present application; Figure 4 A schematic flow chart of a control method for an all-solid-state high repetition rate Q-switched laser provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments of the present application, other embodiments obtained by ordinary technicians in this field without making creative work all belong to the protection scope of the present application.

[0021] In the following, the terms "first", "second", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0022] In addition, in the present application, directional terms such as "upper", "lower", "inner" and "outer" are defined relative to the orientation of the components schematically placed in the drawings. It should be understood that these directional terms are relative concepts. They are used for relative description and clarification, and they can change accordingly according to the changes in the orientation of the components placed in the drawings.

[0023] In order to facilitate the understanding of the technical solution of the application, some existing technologies involved in this application are first explained below.

[0024] All-solid-state high repetition rate Q-switched laser (hereinafter referred to as all-solid-state high repetition rate Q-switched laser) is an advanced laser technology, which is widely used in the fields of spectral analysis, precision measurement, remote sensing, etc. It combines all-solid-state laser, Q-switched technology and wavelength tunable characteristics, and can output laser pulses of specific wavelength at high repetition rate.

[0025] All-solid-state lasers use solid-state gain media to generate lasers. Compared with gas or dye lasers, all-solid-state lasers have the advantages of small size, high efficiency, and good stability.

[0026] Q-switching technology is used to generate high peak power laser pulses. By temporarily introducing loss in the laser cavity (such as using an acousto-optic Q-switcher or an electro-optic Q-switcher), the laser oscillation in the cavity is prevented. When the loss is suddenly removed, the energy is released instantly, generating a narrow pulse width, high peak value laser pulse. Q-switching technology includes acousto-optic Q-switching (AO Q-switching) and electro-optic Q-switching (EO Q-switching). Acousto-optic Q-switching (AO Q-switching) uses the acousto-optic effect to modulate the loss in the laser cavity. Electro-optic Q-switching (EOQ-switching) uses the electro-optic effect to quickly adjust the optical path or loss in the laser cavity.

[0027] Wavelength tunability allows the laser to output different wavelengths within a certain range. This is usually achieved by adding tunable filter elements (such as prisms, gratings, Fabry-Perot cavities, etc.) into the laser cavity. By adjusting these elements, different resonance modes can be selected, thereby changing the wavelength of the laser output.

[0028] All-solid-state high repetition rate Q-switched lasers have a wide range of application backgrounds and important significance in modern science and technology and industry, covering the fields of medical treatment, communication, material processing and environmental monitoring. In the medical field, all-solid-state high repetition rate Q-switched lasers can be used to perform precise tissue cutting and laser cosmetic treatment by selecting specific wavelengths, reducing damage to surrounding tissues and improving treatment effects and safety. In ophthalmic surgery, the use of all-solid-state high repetition rate Q-switched lasers can achieve more precise corneal cutting and enhance the accuracy of vision correction. In the field of communications, all-solid-state high repetition rate Q-switched lasers are the core components of optical fiber communication systems. By adjusting the wavelength, wavelength division multiplexing (WDM) between different channels is achieved, which improves the bandwidth and transmission efficiency of optical fiber communication. All-solid-state high repetition rate Q-switched lasers also have important applications in sensing and measurement systems. By accurately adjusting the wavelength, gas composition, temperature, pressure, etc. are detected, providing data support for environmental monitoring and industrial control. The field of material processing is another important application direction. Laser cutting, welding, marking and other processes rely on the high energy density and precise control of lasers. By adjusting the wavelength, the wavelength suitable for specific materials is selected to achieve efficient and precise processing. All-solid-state high repetition rate Q-switched lasers can achieve high repetition rate pulse output by controlling pulse width and frequency, meeting the needs of high-speed and precision processing. In scientific research, all-solid-state high repetition rate Q-switched lasers are used for spectral analysis, nonlinear optical experiments, laser-induced fluorescence and other research. The wavelength is adjusted to achieve fine spectral analysis of different substances and reveal their internal structure and properties. The short pulse output of all-solid-state high repetition rate Q-switched lasers provides a powerful tool for time-resolved spectroscopy and pump-probe experiments, promoting the development of basic scientific research.

[0029] In summary, all-solid-state high-repetition-rate Q-switched lasers have important application backgrounds and significance in many fields. They not only improve the technical level and production efficiency of various industries, but also provide advanced experimental means for scientific research. In the future, they will show unique advantages and broad application prospects in more fields.

[0030] In all-solid-state high repetition rate Q-switched lasers, birefringent filter groups can be used as wavelength tuning elements. By selecting multiple birefringent filters of appropriate thickness and combining them together, the optical axes can be aligned with each other. By accurately adjusting the optical axis angle of the birefringent filter, the phase difference and interference conditions in the filter group can be controlled, thereby selecting and transmitting lasers of specific wavelengths. In addition, the gain medium of the all-solid-state high repetition rate Q-switched laser is a laser crystal, which can be encapsulated in a TEC module that can be temperature-controlled. By changing the temperature of the laser crystal, the peak wavelength of the crystal gain spectrum will drift, so that the maximum and minimum values ​​of the emission wavelength that can be achieved will change. In all-solid-state high repetition rate Q-switched lasers, electro-optical Q-switched switches can be used as Q-switched elements. By quickly adjusting the Q value of the laser resonant cavity, the output and precise control of pulsed lasers can be achieved in all-solid-state high repetition rate Q-switched lasers. By controlling the modulation voltage, switching frequency, duty cycle and other parameters of the electro-optical Q-switched switch, the pulse intensity, width, repetition frequency and other parameters of the all-solid-state high repetition rate Q-switched laser can be adjusted.

[0031] However, the current all-solid-state high repetition rate Q-switched lasers have the following technical problems: Temperature changes in laser crystals can affect their optical properties and stability. Too high or too low temperatures can cause crystal performance degradation or damage. Uneven temperature distribution inside the crystal can lead to thermal lens effects, affecting the quality of the laser beam. The optical axis angle of the birefringent filter needs to be precisely adjusted. A slight deviation may lead to inaccurate wavelength selection. Mechanical vibrations and position changes in the filter set may affect the stability of wavelength tuning. The laser optical path needs to be precisely aligned to ensure that the light beam can effectively pass through the filter set. The Q-switch needs to respond quickly and work synchronously with the rest of the laser resonant cavity. Any delay or asynchrony may affect the quality of the laser pulse. The modulation voltage of the Q-switch at different wavelengths needs to be precisely controlled to ensure the stability and consistency of the pulsed laser output.

[0032] Therefore, how to quickly and effectively adjust and control the crystal temperature, birefringent filter set, and Q-switching switch module in the all-solid-state high-repetition-rate Q-switched laser to achieve efficient, wide-range wavelength-tuned Q-switched pulse laser output is a technical problem that urgently needs to be solved.

[0033] Figure 1 This is a structural block diagram of the control system of the all-solid-state high repetition rate Q-switched laser provided in an embodiment of the present application.

[0034] In order to solve the above technical problems, combined Figure 1As shown, the embodiment of the present application provides a control system for an all-solid-state high repetition rate Q-switched laser, including: an all-solid-state high repetition rate Q-switched laser 10 and a control module 20. The all-solid-state high repetition rate Q-switched laser 10 is configured to output current operating parameters, and the all-solid-state high repetition rate Q-switched laser 10 includes a TEC module, a Q-switching module, and a wavelength adjustment module; the control module 20 is connected to the all-solid-state high repetition rate Q-switched laser 10 for communication, and the control module 20 is configured to construct a target model; the target model is constructed based on the historical operating parameters of the all-solid-state high repetition rate Q-switched laser 10; the current operating parameters of the all-solid-state high repetition rate Q-switched laser 10 are obtained; based on the target model and the current operating parameters, the TEC module, the Q-switching module, and the wavelength adjustment module are controlled to work.

[0035] By adopting the control system of the all-solid-state high repetition rate Q-switched laser 10 provided in the embodiment of the present application, it is possible to adjust the TEC module, Q-switching module and wavelength adjustment module of the all-solid-state high repetition rate Q-switched laser 10 based on the current operating parameters of the all-solid-state high repetition rate Q-switched laser 10, so as to adjust the beam quality, wavelength and laser pulse of the laser output by the all-solid-state high repetition rate Q-switched laser 10, maintain the stability and consistency of the laser output by the all-solid-state high repetition rate Q-switched laser 10, improve the quality of the output laser, and improve the accuracy of the wavelength of the output laser.

[0036] Figure 2 A schematic diagram of the structure of an all-solid-state high repetition rate Q-switched laser provided in an embodiment of the present application.

[0037] Optionally, combined Figure 2 As shown, the all-solid-state high repetition rate Q-switched laser 10 includes: a pump source 101, a lens coupling system 102, a first cavity mirror 103, a laser crystal 104, a second cavity mirror 105, a wavelength adjustment module 106, a Q-switching module 107, an output mirror 108, a TEC module (not shown in the figure) and a transmission optical fiber 109.

[0038] Specifically, the lens coupling system 102 is configured to collimate, focus and shape the laser emitted by the pump source to obtain a suitable focused light spot, thereby effectively coupling it into the gain medium crystal to improve the optical coupling efficiency.

[0039] The wavelength adjustment module 106 includes a first birefringent filter and a second birefringent filter. By adjusting the angle of the optical axis between the first birefringent filter and the second birefringent filter, the phase difference and interference condition between the first birefringent filter and the second birefringent filter are controlled, thereby selecting and transmitting laser light of a specific wavelength. The first birefringent filter and the second birefringent filter are placed in the optical path of the laser cavity at the Brewster angle and installed in an electric mirror frame that can adjust the rotation angle of the optical axis. In the system, the laser wavelength can be quickly adjusted by adjusting and changing the optical axis angle of the birefringent filter group.

[0040] The Q-switching module 107 is an electro-optical Q-switching switch, which is configured to adjust the pulse intensity, width, repetition frequency and other parameters of the all-solid-state high-repetition-rate Q-switched laser. Specifically, by controlling the modulation voltage, switching frequency and duty cycle of the electro-optical Q-switching switch, the pulse intensity, width, repetition frequency and other parameters of the Q-switched laser can be adjusted.

[0041] Specifically, Figure 2 The arrow in the middle indicates the propagation path of the light. After the light in the sensing optical fiber 109 is output through the output head, it starts to propagate in the direction indicated by the arrow, forming the optical path in the all-solid-state high repetition rate Q-switched laser. The arrow connected to the output mirror 108 indicates the output laser of the all-solid-state high repetition rate Q-switched laser.

[0042] It is understandable that the embodiment of the present disclosure does not specifically limit the installation method between the TEC module and the laser crystal 104, and it is sufficient that the heat generated by the TEC module is transported to the laser crystal 104 through the air duct, so as to achieve the temperature control of the laser crystal 104 through the TEC (Thermoelectric Cooler, semiconductor cooling sheet) module.

[0043] In one embodiment, the laser crystal package is fixed in a brass module and a TEC module is used to precisely control its temperature.

[0044] Figure 3 A schematic diagram of the structure of a control system for an all-solid-state high repetition rate Q-switched laser provided in an embodiment of the present application.

[0045] Optionally, combined Figure 3 As shown, the control module 20 includes: a spectrometer oscilloscope power meter 201, a data acquisition card 202 and a computing device 203. The spectrometer oscilloscope power meter 201 and the data acquisition card 202 are connected in communication, and the data acquisition card 202 and the computing device 203 are connected in communication. The spectrometer oscilloscope power meter 201 includes a receiver 2011. The receiver 2011 is configured to receive the output laser of the all-solid-state high-repetition-rate Q-switched laser, so that the spectrometer oscilloscope power meter 201 receives the output laser and converts it into an electrical signal. The data acquisition card 202 is configured to collect the electrical signal of the output laser in the spectrometer oscilloscope power meter 201 and convert it into a digital signal. The computing device 203 is configured to receive the digital signal (i.e., the current operating parameter) output by the data acquisition card 202, and is configured to control the operation of the TEC module, the Q-switching module and the wavelength adjustment module based on the target model and the current operating parameters.

[0046] Specifically, combined Figure 3 As shown, the computing device 203 outputs control instructions A, B and C to control the operation of the TEC module, the Q-switching module and the wavelength adjustment module respectively.

[0047] It should be understood that the computing device 203 Figure 3 The computing device 203 is shown in the form of a computer, but the specific form of the computing device 203 is not limited. The computing device 203 can also be a mobile phone, a mobile wearable device, a microcomputer, or other device forms that can realize the functions of the above-mentioned computing device 203. This application does not make any limitation here.

[0048] In view of the above technical problems, the present invention proposes a control method for an all-solid-state high repetition rate Q-switched laser to solve the technical problems faced by the all-solid-state high repetition rate Q-switched laser in the process of wavelength tuning. The system can use the spectrum, pulse width, repetition frequency, power and other databases collected by spectrometers, oscilloscopes, power meters, etc. in real time, and combine the crystal temperature-Q-switching switch multi-directional control module to adaptively adjust and control the wavelength tuning process of the all-solid-state high repetition rate Q-switched laser, which can improve the robustness of the control system and increase the wavelength tuning range of the Q-switched pulse laser, thereby reducing the redundant operation of wavelength tuning of the all-solid-state high repetition rate Q-switched laser in the process of realizing intelligent pulse laser output.

[0049] Figure 4 A schematic flow chart of a control method for an all-solid-state high repetition rate Q-switched laser provided in an embodiment of the present application.

[0050] Combination Figure 1 and Figure 3 The control system of the all-solid-state high repetition rate Q-switched laser shown in the embodiment of the present application also provides a control method of the all-solid-state high repetition rate Q-switched laser, the all-solid-state high repetition rate Q-switched laser includes a TEC module, a Q-switching module and a wavelength adjustment module, and the execution body of the method can be a control module, such as Figure 4 The control method of the all-solid-state high repetition rate Q-switched laser shown includes steps S100 to S500, which are specifically as follows: Step S100, constructing a target model; the target model is constructed based on historical operating parameters of the all-solid-state high repetition rate Q-switched laser.

[0051] In this step, a target model is constructed to facilitate the regulation of the all-solid-state high repetition rate Q-switched laser. The target model is based on active feedback learning-self-adjusting convolutional neural network.

[0052] Optionally, the wavelength adjustment module includes a first birefringent filter and a second birefringent filter, and step S100 includes steps S101 to S105, which are specifically as follows: Step S101, obtaining multiple groups of historical operating parameters of the all-solid-state high-repetition-rate Q-switched laser; the historical operating parameters include historical working parameters and historical output parameters corresponding to the historical working parameters; the historical working parameters include historical pump source current, historical crystal temperature, historical Q-switching parameters of the Q-switching module, and historical optical axis angles of the first birefringent filter and the second birefringent filter in the wavelength adjustment module; the historical output parameters include historical output spectrum, historical output pulse width, historical output repetition frequency, and historical output power.

[0053] Exemplarily, the number of groups of historical operating parameters is 10,000 groups.

[0054] Step S102, preprocessing each set of historical operating parameters.

[0055] In this step, the preprocessed historical operating parameters can facilitate the model to learn the correlation between various parameters included in the historical operating parameters of the all-solid-state high repetition rate Q-switched laser.

[0056] Optionally, step S102 includes step S1021 and / or step S1022, which are specifically as follows: Step S1021, standardizing the historical output parameters in the historical operation parameters based on a preset formula.

[0057] In this step, standardization is used to ensure that the data are processed on the same scale and eliminate the impact of dimensional differences.

[0058] Optionally, step S1021 includes step S10211 and step S10212, which are specifically as follows: Step S10211, taking the historical output spectrum, historical output pulse width, historical output repetition frequency, and historical output power in the historical output parameters as the target historical parameters .

[0059] Step S10212, calculating the average value of the target historical parameters among multiple groups of historical operating parameters and standard deviation .

[0060] The preset formulas include: ; is the standardized historical output parameter.

[0061] In this embodiment, the purpose of standardizing the historical operating parameters is to convert different parameter data to a relatively consistent scale. All collected parameters such as spectrum, pulse width, repetition frequency, power data, etc. are processed. The purpose of standardization is to convert these different parameter data to a relatively consistent scale, eliminate dimensional differences, and facilitate subsequent data analysis and model training.

[0062] For example, in the target history parameter When outputting spectra for history, x represents each point in the spectral data or a group of spectral data. When outputting pulse width for history, x represents each pulse width value collected. When outputting repetition frequency for history, x represents the repetition frequency value for each measurement. When it is the historical output power, x represents the output power value collected each time.

[0063] Step S1022: filter the spectrum in the historical output parameters.

[0064] In this step, the spectrum is filtered to remove noise from the spectrum and retain useful signal features, thereby improving the quality and accuracy of the spectral data.

[0065] In this embodiment, the historical operating parameters are preprocessed so that the model can learn the correlation between various parameters included in the historical operating parameters of the all-solid-state high repetition rate Q-switched laser.

[0066] It should be understood that in actual operation, standardization and filtering and denoising are two independent steps, which process data for different needs. Standardization makes all parameter data processed on the same scale, while filtering and denoising focuses on improving the quality of spectral data.

[0067] Specifically, in data preprocessing, filtering technology is also needed to remove noise from the data and retain useful signal features. Commonly used filtering methods include mean filtering and Gaussian filtering, which can effectively smooth the data and reduce the interference of noise on model training.

[0068] Exemplarily, in the case of mean filtering, the signal is smoothed by locally averaging each data point in the signal. The specific process is: select a window size, such as 3 or 5 data points. For each data point, replace it with the mean of the data points in the window around it. This means that the values ​​of all data points in the window will be averaged, and the average value will be used as the filtering result of the current point. The filtering result is a smooth signal, and the sharp fluctuations of the noise are suppressed. In the spectrum, the mean filter will smooth the spectrum curve, remove sharp fluctuations, and thus reduce the impact of noise. This is especially effective when the noise in the spectral data is mainly manifested as high-frequency fluctuations.

[0069] Exemplarily, in the case of Gaussian filtering, a weight window based on Gaussian distribution is used to smooth the signal. Unlike mean filtering, Gaussian filtering assigns different weights to data points according to the distance. The weight of the center point is the largest, and the weight of the point farther from the center is smaller. The specific process is: select a standard deviation σ to define the width of the Gaussian distribution. For each data point, a Gaussian weight function is applied to calculate the weighted average of the point and its neighborhood, and the weight is determined by the Gaussian distribution. The filtering result is a smoother signal, while retaining the overall trend of the original signal, and the noise is effectively suppressed. Gaussian filtering can effectively smooth the curve in spectral processing and remove noise while retaining important signal features. It is particularly effective when processing spectral data with Gaussian noise characteristics.

[0070] In summary, mean filtering is more suitable for removing relatively uniform high-frequency noise, but it may cause excessive smoothing of the signal and loss of details. Gaussian filtering effectively removes noise while retaining signal details, and is more suitable for spectral signal processing that needs to retain more details.

[0071] Step S103: Generate an agent identifier based on active feedback learning technology and historical operation parameters.

[0072] Optionally, step S103 includes step S1031, which is specifically as follows: Step S1031, based on the difference-driven learning technology, construct multiple groups of positively correlated sample groups and negatively correlated sample groups of historical operating parameters; the proxy identifier includes the positively correlated sample group and the negatively correlated sample group; the positively correlated sample group and the negatively correlated sample group are both obtained by transforming the historical operating parameters.

[0073] In this embodiment, a positive correlation sample group and a negative correlation sample group are constructed to expand the training data set and improve the generalization ability of the model.

[0074] Optionally, the loss function for difference-driven learning is: ; in, is the agent identifier, For sample The feature mapping function is For sample The feature map of For sample The feature map of is the Euclidean distance between positively correlated sample groups, is the Euclidean distance between negatively correlated sample groups, is the preset boundary parameter, is the total number of positively correlated sample groups and negatively correlated sample groups.

[0075] In this embodiment, a proxy identifier is generated by difference-driven learning, and the model is trained in an unsupervised manner to learn effective feature representations, thereby realizing intelligent control of an all-solid-state high-repetition-rate Q-switched laser. The loss function in active feedback learning is used to train the model so that the feature representations of samples of the same type are as close as possible, and the feature representations of samples of different types are as far apart as possible. This improves the accuracy of the generated proxy identifier, enhances the feature learning ability of the model, and provides an optimization target.

[0076] Specifically, active feedback learning technology includes difference-driven learning technology.

[0077] Specifically, if the agent identifier , then it means the sample and is a similar pair (positively correlated sample group), that is, the same input sample of the same category or after data enhancement. , then it means the sample and are non-similar pairs (negatively correlated sample groups), that is, samples of different categories.

[0078] For sample The feature mapping function of Embedded representation in feature space. The purpose of the model is to learn a good feature representation. , making the distance between similar samples small and the distance between dissimilar samples large.

[0079] For sample The feature map is usually a sample A positive sample of Similar or identical samples. Positive samples can be samples The result obtained by performing data augmentation (such as translation, rotation, scaling, etc.).

[0080] For sample The feature map represents the sample Dissimilar negative samples. Negative samples are usually The purpose of the samples of different categories is to separate their representations as much as possible in the feature space.

[0081] is the square of the Euclidean distance between positively correlated sample groups. As the first term of the loss function, the goal is to minimize the distance between similarly correlated sample groups so that positively correlated sample groups can be clustered together.

[0082] is the square of the Euclidean distance between negatively correlated sample groups. As the second term of the loss function, the goal is to maximize the distance between non-similar correlated sample groups.

[0083] is a preset boundary parameter used to ensure that the distance between negatively correlated sample groups is greater than this value. The "maximization" part of the loss function uses this parameter, which means that if the distance between negatively correlated sample groups is less than , a penalty term will be generated.

[0084] is the total number of positively correlated sample groups and negatively correlated sample groups, usually used to calculate the average loss.

[0085] Specifically, the loss function of difference-driven learning aims to enable the model to automatically generate high-quality proxy identifiers, thereby improving feature representation learning through active feedback learning. Specifically, for similar related sample groups (positively related sample groups, ), we hope that their distance in the feature space is as small as possible. For dissimilar related sample groups (negatively related sample groups, ), and we hope that their distance is at least In this way, the model can autonomously learn the structure and patterns in the data without explicit annotations, thereby generating effective feature representations.

[0086] Exemplarily, positively correlated sample groups are composed of different versions of the same data sample. It can be understood that they are generated from the same sample through data augmentation or data transformation, so their generation process is based on the transformation of the real sample. When constructing related sample group pairs, the same original data sample (such as an image or signal) is usually subjected to some kind of data augmentation or transformation, such as rotation, scaling, cropping, color perturbation, etc. These transformations do not change the essential properties of the sample, but give a different perspective or representation. Therefore, positively correlated sample groups are different representations generated from the same data sample, and the model should learn that they represent the same category or have the same semantic information. For example, given an image , two different transformation versions can be obtained through data enhancement and . These two samples form a positively correlated sample group because they come from the same original image.

[0087] The negatively correlated sample group is composed of different data samples. It can be understood that it is not generated by simulation, but different samples selected directly from the training data set. The generation of the negatively correlated sample group depends on the diversity and randomness of the data set itself. In difference-driven learning, different data samples usually refer to two samples randomly selected from the training set that have no direct relationship. The samples in the negatively correlated sample group usually come from different original data (such as different images or signals). These samples are semantically or categorically unrelated, so the model should learn the differences between them and separate them as much as possible in the feature space. For example, suppose there are two different original image samples and , these samples are completely different, and they can constitute a negatively correlated sample group.

[0088] Distinguish between positively correlated sample groups and negatively correlated samples: The positively correlated sample group is a transformed version: The two parts of the positively correlated sample group (such as and ) is composed of the same original data sample (such as ) are generated by different data augmentation strategies (such as rotation, scaling, etc.), they should represent the same semantics or categories. The negatively correlated sample group is a different data sample: the two parts of the negatively correlated sample group come from completely different original data samples (such as and ), which are semantically or categorically different.

[0089] In the above description, "different data samples" usually refer to two raw data samples that have no direct relationship (such as two random sets of historical operating parameters). These samples have no similarity or semantic connection and are usually randomly selected in the training data set. For negatively correlated sample groups, the model should learn to disperse their feature representations as much as possible in the feature space to enhance the model's ability to distinguish different categories.

[0090] In difference-driven learning, "positively correlated sample groups" are different versions of the same original sample generated through transformation, and "negatively correlated sample groups" are different original samples selected from the training dataset. Therefore, the similarities of positively correlated sample groups should be captured by the model, while the differences of negatively correlated sample groups should be distinguished by the model.

[0091] Specifically, the loss function of difference-driven learning can minimize the loss function, and the model can transform the input data The feature representation The corresponding agent identifier Keep it consistent. The proxy identifier reflects the model's understanding of the internal structure of the data. Minimizing the loss function helps optimize the model's feature representation so that it more accurately reflects the true distribution and pattern of the data and improves the accuracy of the generated proxy identifier.

[0092] Furthermore, the loss function of difference-driven learning guides the model's parameter update so that similar data (defined by proxy identifiers) are close to each other in the feature space, while dissimilar data remain separated. In this way, the model can automatically learn how to effectively distinguish different types of input data without explicit labels, helping the model learn more meaningful feature representations to enhance the model's feature learning capabilities.

[0093] Furthermore, in active feedback learning, there are no traditional explicit labels to directly optimize the model, so a suitable optimization objective needs to be defined. This loss function provides such a goal: by minimizing the value of the loss function, the model is guided to continuously adjust its parameters to learn better and more general feature representations.

[0094] Finally, in the all-solid-state high repetition rate Q-switched laser control method, active feedback learning works in conjunction with a self-adjusting convolutional neural network. This loss function is used to process the large amount of data output from the all-solid-state high repetition rate Q-switched laser, and to generate proxy identifiers through difference-driven learning to expand the training data set. Optimizing this loss function enables the self-adjusting convolutional neural network to better identify and control the characteristics of the laser output parameters, such as wavelength, pulse width, and power, to achieve intelligent control.

[0095] Step S104: adding the agent identifier and multiple sets of historical operating parameters to the training data set.

[0096] In this step, the training dataset is expanded by adding proxy identifiers to the training dataset, thereby increasing the size and diversity of the training dataset. By introducing more sample pairs, the model can learn richer feature relationships, avoid overfitting, and enhance generalization capabilities.

[0097] Specifically, the expanded training dataset contains not only the original acquired data, but also the proxy identifiers generated by the active feedback learning technique. This means that the training dataset has added more data samples through the proxy identifiers on the basis of the original real data. These newly added samples may be different transformed versions of the original data, or they may be new data points generated by the active feedback learning technique.

[0098] Exemplarily, the training data set includes: multiple sets of preprocessed historical operating parameters and proxy identifiers. The generated proxy identifiers are usually similar to the preprocessed data. The preprocessing process (such as standardization, filtering, etc.) ensures the consistency and quality of the data, and the generation of proxy identifiers increases the diversity of the data on this basis. Therefore, the proxy identifiers are consistent in form with the original preprocessed data and can be seamlessly integrated into the training data set, helping the self-adjusting convolutional neural network learn a wider range of features.

[0099] Step S105, using the training data set to train the self-adjusting convolutional neural network to determine the target model.

[0100] Optionally, step S105 includes steps S1051 to S1056, which are specifically as follows: Step S1051, constructing a self-adjusting convolutional neural network; the self-adjusting convolutional neural network includes a convolutional layer, a pooling layer, and a fully connected layer.

[0101] Step S1052, input the training data set into the self-adjusting convolutional neural network.

[0102] Step S1053, the convolution layer extracts feature maps from the data based on multi-level convolution operations.

[0103] In this step, the main function of the convolution layer is to extract local features in the input data (i.e., the feature map mentioned above). These features reflect the details and changes of the laser output, including but not limited to local spectral features, pulse shape features, power features, and repetition frequency features. The convolution layer scans these features through a series of convolution kernels (filters) to extract local information at different scales and positions, thereby providing rich feature expressions for subsequent model layers.

[0104] Specifically, local features of the spectrum: such as peak position, peak height, absorption line, edge changes, small fluctuations in the spectrum, etc. These features can reveal the wavelength adjustment range, spectral width and stability of the laser.

[0105] Pulse shape characteristics: including the rising edge, falling edge, peak width and position of the pulse. These characteristics are crucial to the intensity and stability of the laser pulse output and reflect the pulse dynamic behavior of the laser.

[0106] Power characteristics: such as short-term fluctuations in output power, power peaks and average values, etc. These characteristics are important for understanding the stability and response characteristics of the laser.

[0107] Repetition frequency characteristics: Slight changes in repetition frequency, stability, and jitter, which can be used to determine whether the operating frequency of the laser meets expectations.

[0108] In step S1054, the pooling layer converts the first size of the feature map into a second size based on a downsampling operation, where the second size is smaller than the first size.

[0109] In this step, through the pooling operation, the size of the feature map is reduced and the computational complexity is reduced, while the most important information for feature extraction and prediction is retained, thereby improving the efficiency and robustness of the self-regulating convolutional neural network model.

[0110] Step S1055, extracting feature information corresponding to the feature map.

[0111] Step S1056: The fully connected layer implements the final classification and regression of feature information based on linear transformation.

[0112] In this step, "final classification and regression of feature information" means that the network completes two different tasks based on the extracted features: final classification of features and final regression of features.

[0113] Specifically, the final classification of the features is defined as follows: The goal of the classification task is to assign the input data to a predefined category. In the context of an all-solid-state high-repetition-rate Q-switched laser control system, the task of feature classification may involve identifying the state or category of the laser under different operating conditions. For example, a self-adjusting convolutional neural network can classify the state of the current laser output into a specific wavelength, pulse width, repetition frequency range, etc. based on the extracted spectral features. The purpose is: The results of the classification can be used for real-time decision-making and control. For example, when a certain abnormal state is detected (such as the output wavelength deviates from the expected value), the control system can adjust the relevant operating parameters to correct the state.

[0114] Specifically, the final regression of the feature is defined as follows: The goal of the regression task is to predict one or more continuous values. In the laser control system, this means predicting the value of certain continuous variables, such as the specific values ​​of output power, pulse width, frequency, etc., based on the extracted feature data. The purpose is: The results of the regression can help accurately adjust the laser parameters to achieve the expected performance. For example, the regression model can predict the output power at different pump source currents or crystal temperatures, so that the system can automatically adjust these parameters to achieve the optimal output state.

[0115] For example, in a self-tuning convolutional neural network, the final classification and regression of features represent two different prediction tasks: classification is used to identify the category or state to which the data belongs, and regression is used to predict the specific numerical output of the data. In the embodiment of the present application, these two tasks help the system to accurately identify and control the output state, thereby optimizing performance and stability.

[0116] Optionally, when implementing the final regression of the features in step S1056, the loss function adopted is mean square error (MES).

[0117] The formula for mean square error is: ;in, is the actual value, is the predicted value, is the number of samples.

[0118] In this embodiment, the self-adjusting convolutional neural network extracts complex patterns and features in the data through multi-level convolution operations, improves the expressive power of the model, and realizes the learning of the operating parameters of the all-solid-state high-repetition-rate Q-switched laser.

[0119] Specifically, complex patterns and features include: interactions between multi-dimensional parameters, complex dynamics in time series, nonlinear features in high-dimensional space, rare or abnormal patterns, and multi-scale features.

[0120] The interactive relationship between multi-dimensional parameters includes: the relationship between spectrum and power and the interaction between pulse width and repetition frequency.

[0121] Relationship between spectrum and power: There is a complex nonlinear relationship between the output spectrum and power of all-solid-state high repetition rate Q-switched lasers. As the power of all-solid-state high repetition rate Q-switched lasers changes, the characteristics of the spectrum (such as central wavelength, bandwidth, peak intensity) may change significantly. This relationship may not only be a simple linear increase or decrease, but also involve more complex patterns, such as the spectrum may have a multi-peak structure in certain power ranges, or the spectral bandwidth may expand nonlinearly with increasing power. Self-adjusting convolutional neural networks are able to capture these complex change patterns through multi-layer convolution operations and learn how to extract features related to power changes from spectral data. Interaction between pulse width and repetition frequency: There is also a complex interaction between pulse width and repetition frequency. When the repetition frequency of an all-solid-state high repetition rate Q-switched laser increases, the pulse width may become narrower, but under certain conditions, nonlinear interactions may occur, resulting in abnormal changes in pulse width. Self-adjusting convolutional neural networks are able to analyze the change patterns of the two under different conditions, identify how the pulse width affects the repetition frequency of the laser in certain specific cases, and thus extract these important influence patterns.

[0122] The complex dynamics in the time series include: the time dependence of the wavelength tuning process and the temporal relationship of the power stability with the noise.

[0123] Time dependence in the wavelength tuning process: The wavelength tuning of a laser is usually not instantaneous, but a dynamic process that changes over time. During the tuning process, the characteristics of the spectrum may gradually change over time, such as the gradual shift of the center wavelength or the evolution of the spectral shape. By analyzing the time series of the data, the self-adjusting convolutional neural network can capture these characteristic patterns that evolve over time, thereby accurately modeling the wavelength tuning process. The temporal relationship between power stability and noise: During the operation of the laser, the output power may show periodic fluctuations or be affected by random noise. These fluctuations and noise are not simple white noise, but may have complex time correlations, such as periodic oscillations in power within a specific time period. The self-adjusting convolutional neural network can identify the periodic patterns of power fluctuations or the characteristics of random noise by analyzing these time series data, thereby improving the stability analysis and control accuracy of the laser.

[0124] The nonlinear characteristics in high-dimensional space include: the influence of nonlinear optical effects and multi-parameter coupling effects.

[0125] Influence of nonlinear optical effects: Nonlinear optical effects inside the laser (such as four-wave mixing, Raman scattering, etc.) may have a significant impact on the shape and stability of the spectrum. These effects usually do not appear as simple linear responses, but have complex, high-dimensional characteristics. For example, nonlinear effects may cause new peaks to appear in the spectrum or cause existing peaks to shift and broaden. Self-adjusting convolutional neural networks can capture these complex nonlinear features in high-dimensional space through multi-layer convolution kernels and integrate them into the performance model of the laser.

[0126] Multi-parameter coupling effects: There may be complex coupling effects between multiple operating parameters of the laser (such as pump current, crystal temperature, modulation voltage of the electro-optical Q-switch, frequency, duty cycle, etc.). For example, changes in pump current may not only affect the output power, but may also indirectly affect the spectral characteristics by changing the temperature of the laser. Self-adjusting convolutional neural networks can identify these complex coupling effects by analyzing these multi-parameter data, thereby establishing a more comprehensive laser performance prediction model.

[0127] Rare or unusual modes include: unusual wavelength drift and power flicker phenomena.

[0128] Abnormal wavelength drift: Under certain specific operating conditions, the wavelength of the laser may drift abnormally, which may be caused by equipment aging, environmental changes, or other uncommon factors. Such abnormal patterns are usually hidden in a large amount of normal operating data and are difficult to be identified by traditional analysis methods. Self-adjusting convolutional neural networks can identify these rare abnormal patterns by learning from a large amount of training data, and effectively detect and process them in practical applications.

[0129] Power flicker phenomenon: In some cases, the output power of the laser may flicker abnormally for a short time. These flickers may be the manifestation of some unstable factors inside the laser or external interference. The self-adjusting convolutional neural network can extract these short-term abnormal features by analyzing subtle changes in the data and identify the potential reasons behind them, thus providing a more accurate diagnostic tool.

[0130] Multi-scale features include: the combination of details and overall patterns and multi-scale analysis of pulse shapes.

[0131] Combination of details and overall patterns: In the spectral data of lasers, there are both microscopic details (such as small peaks or tiny fluctuations in the spectrum) and macroscopic overall trends (such as the overall shape and distribution of the entire spectrum). Self-adjusting convolutional neural networks can capture these details and overall patterns at the same time through convolution kernels at different levels. For example, the convolution kernel can focus on detailed features at a small scale while capturing the overall shape of the spectrum at a large scale. Such multi-scale feature extraction helps to fully understand the spectral characteristics of the laser.

[0132] Multi-scale analysis of pulse shape: The shape of a laser pulse may contain multiple different time-scale features, such as the pulse's rise time, peak position, fall time, and the length of its tail. These features may reflect different operating states or internal mechanisms of the laser. Self-tuning convolutional neural networks can analyze these pulse shapes at different time scales to fully grasp the multi-dimensional changes in pulse characteristics.

[0133] In summary, complex patterns and features cover multi-dimensionality, nonlinearity, time dependence, rare events, and multi-scale analysis in laser data. These features are often difficult to extract and analyze through simple statistical or linear models, and self-adjusting convolutional neural networks can effectively capture and learn these complex features through their multi-level and multi-scale convolution operations, thereby greatly improving the model's understanding and prediction capabilities for laser systems. The use of this deep learning model can significantly improve the performance optimization, stability analysis, and fault detection of lasers.

[0134] In this embodiment, multiple sets of historical operating parameters of the all-solid-state high repetition rate Q-switched laser are obtained to build a target model based on the historical operating parameters of the all-solid-state high repetition rate Q-switched laser, so as to facilitate determining how to regulate the operation of the all-solid-state high repetition rate Q-switched laser according to the target model and the current operating parameters, thereby realizing the regulation of the output laser of the all-solid-state high repetition rate Q-switched laser.

[0135] Optionally, after step S1056, step S1057 is further included, which is specifically as follows: Step S1057, optimizing the self-adjusting convolutional neural network based on a dynamic learning rate optimization algorithm or a performance optimization algorithm.

[0136] In this embodiment, the self-adjusting convolutional neural network is optimized by a dynamic learning rate optimization algorithm, which can improve the convergence speed and stability of the training. The self-adjusting convolutional neural network is optimized by a performance optimization algorithm, which can improve the performance of the model, prevent the model from overfitting, and improve the adaptability and real-time performance of the model.

[0137] Optionally, optimizing the self-adjusting convolutional neural network based on a dynamic learning rate optimization algorithm includes: calculating predicted values ​​and loss values ​​using forward propagation; calculating gradients through back propagation; and updating model parameters using a dynamic learning rate optimization algorithm.

[0138] Optionally, the update formula of the dynamic learning rate optimization algorithm is: ; in, are the model parameters of the self-tuning convolutional neural network, is the learning rate, is the momentum estimate of the gradient, is the second moment estimate of the gradient, is a constant to prevent division by zero.

[0139] Specifically, the training data is input into the self-adjusting convolutional neural network model, the prediction value and loss value are calculated by forward propagation, the gradient is calculated by back propagation, and the model parameters are updated using the dynamic learning rate optimization algorithm. During the training process, it is necessary to continuously adjust the hyperparameters (such as learning rate, batch size, etc.) to improve the training effect of the model.

[0140] Exemplarily, the dynamic learning rate optimization algorithm is a method for model parameter optimization. It is mainly used to update the weights and biases of the neural network during the training process to minimize the loss function value of the model, thereby improving the model's fitting ability. The dynamic learning rate optimization algorithm speeds up the training speed and improves the stability of the model by calculating the adaptive learning rate of each parameter. The working level of the dynamic learning rate optimization algorithm is: working at the micro-optimization level, that is, in each iteration (each training step), it dynamically adjusts the learning rate of each parameter, and determines the direction and step size of the parameter update based on the exponential weighted average of the previous gradients and the square of the gradients.

[0141] Optionally, performance optimization algorithms include: hyperparameter tuning, regularization techniques, model evaluation, and online learning. Hyperparameter tuning: Use methods such as cross-validation and grid search to find the optimal hyperparameter combination to ensure that the model training process is more efficient. Regularization techniques: Add regularization terms (such as L2 regularization) to the loss function or use techniques such as Dropout to prevent the model from overfitting on the training data. Model evaluation: Use evaluation indicators on the validation set to monitor model performance and adjust hyperparameters or model structure to improve model effects. Online learning: Learn by continuously inputting new data and updating model parameters so that the model maintains its adaptability to new data in actual operation.

[0142] Exemplarily, the performance optimization algorithm is used for higher-level model optimization to ensure that the overall structure and performance of the model are optimal. The specific goals are as follows: The goal of hyperparameter tuning is to adjust hyperparameters such as learning rate, batch size, network depth, etc. to optimize the model training effect. The goal of regularization technology is to prevent overfitting and improve the generalization ability of the model. The goal of model evaluation is to ensure that the model performs optimally on the validation set to prevent overfitting or underfitting. The goal of online learning is to maintain its adaptability to new data by updating the model in real time. The working level of the performance optimization algorithm is: working at the macro optimization level, involving the global strategy and design of model training, which is a higher-level optimization. At multiple stages of model training, these techniques are interrelated and cyclically alternating to continuously improve the overall performance and stability of the model.

[0143] The dynamic learning rate optimization algorithm and the performance optimization algorithm are both different and related. The difference is referred to the above introduction, and this application will not repeat it here. The connection is as follows: The connection between hyperparameter tuning and dynamic learning rate optimization algorithm: The learning rate directly affects the effect of the dynamic learning rate optimization algorithm; during the hyperparameter tuning stage, the learning rate and other parameters of the dynamic learning rate optimization algorithm can be adjusted to obtain better training results.

[0144] The connection between regularization technology and dynamic learning rate optimization algorithm: Regularization directly affects the value of the loss function, and the dynamic learning rate optimization algorithm uses this regularized loss function when calculating the gradient, which indirectly affects the parameter update.

[0145] The connection between model evaluation and dynamic learning rate optimization algorithm: Model evaluation can evaluate the effect of the model optimized by the dynamic learning rate optimization algorithm after multiple iterations. If the effect is not good, it is necessary to adjust the hyperparameters or regularization strategy.

[0146] The connection between online learning and dynamic learning rate optimization algorithm: Online learning requires the model to continuously accept new data and update, which is usually combined with a dynamic learning rate optimization algorithm to achieve real-time parameter updates.

[0147] In summary, the two are interconnected and work together to improve the performance and adaptability of the model. It can also be understood that the dynamic learning rate optimization algorithm is one of the specific tools to implement these higher-level strategies.

[0148] Step S200, obtaining current operating parameters of the all-solid-state high repetition rate Q-switched laser.

[0149] In this step, the gap between the current operating parameters and the desired operating parameters can be determined based on the acquired current operating parameters, thereby facilitating the regulation of the output laser of the all-solid-state high repetition rate Q-switched laser.

[0150] Specifically, the current operating parameters include: the current operating parameters include the current working parameters and the current output parameters corresponding to the current working parameters; the current working parameters include the current pump source current, the current crystal temperature, the current Q-switching parameters of the Q-switching module, and the current optical axis angles of the first birefringent filter and the second birefringent filter in the wavelength adjustment module; the current output parameters include the current output spectrum, the current output pulse width, the current output repetition frequency, and the current output power.

[0151] Step S300, preprocessing current operating parameters.

[0152] In this step, the current operating parameters are preprocessed so that the target model can better determine the current working condition of the all-solid-state high repetition rate Q-switched laser according to the current operating parameters.

[0153] Specifically, the implementation method of step S300 refers to the method shown in step S102, and this application will not go into details here.

[0154] Step S400, based on the target model and the current operating parameters, determine a first working instruction corresponding to the TEC module, a second working instruction corresponding to the Q-switching module, and a third working instruction corresponding to the wavelength adjustment module.

[0155] In this step, the first work instruction is Figure 3 The second working instruction is C shown in Figure 3 The third working instruction is Figure 3 The instruction shown in B.

[0156] Specifically, the preprocessed current operating parameters are input into the target model, and the target model outputs a first working instruction corresponding to the TEC module, a second working instruction corresponding to the Q-switching module, and a third working instruction corresponding to the wavelength adjustment module.

[0157] Step S500, controlling the TEC module to work according to the first working instruction, controlling the Q-switching module to work according to the second working instruction, and controlling the wavelength adjustment module to work according to the third working instruction.

[0158] Specifically, the first working instruction, the second working instruction and the third working instruction can be direct working instructions or indirect working instructions. The direct working instruction is an instruction for directly controlling the current and voltage of the hardware, and the indirect working instruction is an instruction for giving the parameter control value, such as adjusting the temperature value or adjusting the optical axis angle value.

[0159] Exemplarily, the first working instruction includes but is not limited to the adjustment temperature value of the TEC module (indirect instruction) or the current or voltage change of the specific hardware (direct instruction). The present application does not limit this, as long as it can adjust the temperature of the laser crystal. The second working instruction includes but is not limited to the modulation voltage, switching frequency and duty cycle to adjust the pulse of the laser output laser. The third working instruction includes but is not limited to the optical axis adjustment parameter between the first birefringent filter and the second birefringent filter (direct instruction) or the wavelength adjustment size instruction (indirect instruction). The present application does not limit this, as long as it can adjust the wavelength.

[0160] The second working instruction is explained. Among them, the modulation voltage, switching frequency and duty cycle can all affect the pulse and spectrum. Specifically, the modulation voltage determines the modulation depth of the electro-optical Q-switching switch, which in turn affects the Q value variation amplitude of the laser. A higher modulation voltage can more effectively reduce the Q value of the resonant cavity, resulting in a stronger pulse output. A lower modulation voltage will produce weaker pulses and may even cause the pulse energy to drop. Changes in the modulation voltage will affect the pulse energy and duration of the laser, which in turn affects the width of the spectrum. A higher modulation voltage usually produces a narrower spectral linewidth, while a lower modulation voltage may cause the spectrum to be broadened.

[0161] The switching frequency directly determines the repetition rate of the pulses. A higher switching frequency will increase the repetition rate of the pulses, but may cause the energy of each pulse to decrease. Conversely, a lower switching frequency will produce fewer pulses, but the energy of each pulse will be higher. The change in switching frequency affects the modulation characteristics of the spectrum. A higher frequency may cause the spectrum to appear broadened in modulation bandwidth, while a lower frequency will make the spectrum more concentrated.

[0162] The duty cycle determines the ratio of the laser's Q-switch on time to the off time. A larger duty cycle will result in an increase in pulse width, which affects the peak power of the pulse. A smaller duty cycle will produce narrower, sharper pulses. Changes in the duty cycle directly affect the width of the pulse, which in turn affects the width of the spectral line. A larger duty cycle may lead to a broadening of the spectral linewidth, while a smaller duty cycle helps to maintain a narrow spectral linewidth.

[0163] In summary, the modulation voltage, switching frequency and duty cycle of the electro-optical Q-switch will jointly affect the pulse energy, width, repetition frequency and other parameters of the Q-switched laser, thereby affecting the line width, central wavelength and modulation characteristics of the output spectrum.

[0164] The control method of the all-solid-state high repetition rate Q-switched laser provided in the embodiment of the present application can adjust the TEC module, Q-switching module and wavelength adjustment module of the all-solid-state high repetition rate Q-switched laser based on the current operating parameters of the all-solid-state high repetition rate Q-switched laser, thereby adjusting the beam quality, wavelength and laser pulse of the laser output by the all-solid-state high repetition rate Q-switched laser, realizing intelligent wavelength and pulse control, efficient dynamic adjustment and optimization, and enhancing the stability and adaptability of the output laser. Maintain the stability and consistency of the laser output by the all-solid-state high repetition rate Q-switched laser, improve the quality of the output laser, and improve the accuracy of the wavelength of the output laser.

[0165] Intelligent wavelength and pulse control: Based on active feedback learning and self-adjusting convolutional neural network technology, this control system can intelligently adjust the birefringent filter to achieve precise wavelength selection and broaden the gain spectrum width by adjusting the crystal temperature. This intelligent control not only improves the accuracy and flexibility of wavelength tuning, but also expands the range of adjustable wavelengths to meet diverse application needs.

[0166] Efficient dynamic adjustment and optimization: By synchronously controlling the modulation voltage, switching frequency, duty cycle and other parameters of the electro-optical Q-switching switch module, the system achieves efficient dynamic adjustment of the pulse width, repetition frequency and power of the Q-switched laser. Compared with traditional control methods, the system can quickly respond to environmental changes and operational requirements, provide more stable and efficient laser output, and improve overall work efficiency.

[0167] Enhanced stability and adaptability: Using active feedback learning technology, the system can continuously learn and optimize its own performance, automatically adapting to different working conditions and environmental changes. This not only improves the long-term stability and reliability of the laser, but also extends the service life of the equipment. The system's adaptive ability ensures excellent performance in a variety of complex application scenarios, reducing the need for manual intervention and maintenance.

[0168] Corresponding to the embodiment of the control method of the all-solid-state high repetition rate Q-switched laser, the present application also provides an embodiment of the control system of the all-solid-state high repetition rate Q-switched laser. The control system of the all-solid-state high repetition rate Q-switched laser includes: an all-solid-state high repetition rate Q-switched laser, configured to output current operating parameters, the all-solid-state high repetition rate Q-switched laser includes a TEC module, a Q-switched module and a wavelength adjustment module; a control module, which is connected to the all-solid-state high repetition rate Q-switched laser for communication, and the control module is configured to build a target model; the target model is built based on the historical operating parameters of the all-solid-state high repetition rate Q-switched laser; the current operating parameters are obtained; the current operating parameters are pre-processed; the current operating parameters are input into the target model, and the target model outputs a first working instruction corresponding to the TEC module, a second working instruction corresponding to the Q-switched module and a third working instruction corresponding to the wavelength adjustment module; the TEC module is controlled to work according to the first working instruction, the Q-switched module is controlled to work according to the second working instruction, and the wavelength adjustment module is controlled to work according to the third working instruction.

[0169] The control system of the all-solid-state high repetition rate Q-switched laser provided by the embodiment of the present application can adjust the TEC module, Q-switching module and wavelength adjustment module of the all-solid-state high repetition rate Q-switched laser based on the current operating parameters of the all-solid-state high repetition rate Q-switched laser, thereby adjusting the beam quality, wavelength and laser pulse of the laser output by the all-solid-state high repetition rate Q-switched laser, realizing intelligent wavelength and pulse control, efficient dynamic adjustment and optimization, and enhancing the stability and adaptability of the output laser. Maintain the stability and consistency of the laser output by the all-solid-state high repetition rate Q-switched laser, improve the quality of the output laser, and improve the accuracy of the wavelength of the output laser.

[0170] It should be noted that those skilled in the art will easily think of other embodiments of the present application after considering the specification and practicing the application disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary technical means in the art that are not disclosed in the present application. The specification and examples are only regarded as exemplary, and the true scope of the present application is indicated by the claims.

[0171] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A control method for an all-solid-state high repetition rate Q-switched laser, characterized in that: The all-solid-state high repetition rate Q-switched laser comprises a TEC module, a Q-switched module and a wavelength adjustment module. The control method of the all-solid-state high repetition rate Q-switched laser comprises: Constructing a target model; the target model is constructed based on historical operating parameters of the all-solid-state high repetition rate Q-switched laser; Obtaining current operating parameters of the all-solid-state high repetition rate Q-switched laser; Preprocessing the current operating parameters; Based on the target model and the current operating parameters, determining a first working instruction corresponding to the TEC module, a second working instruction corresponding to the Q-switching module, and a third working instruction corresponding to the wavelength adjustment module; The TEC module is controlled to work according to the first working instruction, the Q-switching module is controlled to work according to the second working instruction, and the wavelength adjustment module is controlled to work according to the third working instruction.

2. The control method of the all-solid-state high repetition rate Q-switched laser according to claim 1, characterized in that: The wavelength adjustment module includes a first birefringent filter and a second birefringent filter, and the target model is constructed, including: Acquire multiple groups of historical operating parameters of the all-solid-state high-repetition-rate Q-switched laser; the historical operating parameters include historical working parameters and historical output parameters corresponding to the historical working parameters; the historical working parameters include historical pump source current, historical crystal temperature, historical Q-switching parameters of the Q-switching module, and historical optical axis angles of the first birefringent filter and the second birefringent filter in the wavelength adjustment module; the historical output parameters include historical output spectrum, historical output pulse width, historical output repetition frequency, and historical output power; preprocessing each set of the historical operating parameters; generating an agent identifier based on active feedback learning techniques and the historical operating parameters; adding the agent identifier and the plurality of sets of the historical operating parameters to a training data set; The training data set is used to train a self-adjusting convolutional neural network to determine the target model.

3. The control method of the all-solid-state high repetition rate Q-switched laser according to claim 2, characterized in that: The preprocessing of each group of historical operating parameters includes: Based on a preset formula, standardize the historical output parameters in the historical operating parameters; and / or, The spectrum in the historical output parameters is subjected to filtering processing.

4. The control method of the all-solid-state high repetition rate Q-switched laser according to claim 3, characterized in that: Standardizing the historical output parameters in the historical operating parameters includes: The historical output spectrum, the historical output pulse width, the historical output repetition frequency, and the historical output power in the historical output parameters are sequentially used as target historical parameters. ; Calculate the average value of the target historical parameters among multiple groups of historical operating parameters and standard deviation ; The preset formula includes: ； Among them, the is the standardized historical output parameter.

5. The control method of the all-solid-state high repetition rate Q-switched laser according to claim 2, characterized in that: The generating of the agent identifier based on the active feedback learning technology and the historical operation parameters includes: Based on the difference-driven learning technology, multiple groups of positively correlated sample groups and negatively correlated sample groups of the historical operating parameters are constructed; the proxy identifier includes the positively correlated sample group and the negatively correlated sample group; the positively correlated sample group and the negatively correlated sample group are both obtained by transforming the historical operating parameters.

6. The control method of the all-solid-state high repetition rate Q-switched laser according to claim 5, characterized in that: The loss function of the difference-driven learning is: ； Among them, the is the agent identifier, the For sample The feature mapping function is For sample The feature map of For sample The feature map of is the Euclidean distance between the positively correlated sample groups, is the Euclidean distance between the negatively correlated sample groups, is the preset boundary parameter, is the total number of the positive correlation sample group and the negative correlation sample group.

7. The control method of the all-solid-state high repetition rate Q-switched laser according to claim 2, characterized in that: The using the training data set to train the self-adjusting convolutional neural network to determine the target model includes: Constructing the self-adjusting convolutional neural network; the self-adjusting convolutional neural network includes a convolutional layer, a pooling layer, and a fully connected layer; Inputting the training data set into the self-tuning convolutional neural network; The convolution layer extracts feature maps from the data based on multi-level convolution operations; The pooling layer converts the first size of the feature map into a second size based on a downsampling operation, where the second size is smaller than the first size; Extracting feature information corresponding to the feature map; The fully connected layer implements the final classification and regression of the feature information based on linear transformation.

8. The control method of the all-solid-state high repetition rate Q-switched laser according to claim 7, characterized in that: After the fully connected layer realizes the final classification and regression of the feature information based on linear transformation, it also includes: The self-adjusting convolutional neural network is optimized based on a dynamic learning rate optimization algorithm or a performance optimization algorithm.

9. The control method of the all-solid-state high repetition rate Q-switched laser according to claim 8, characterized in that: The update formula of the dynamic learning rate optimization algorithm is: ； Among them, the are the model parameters of the self-adjusting convolutional neural network, is the learning rate, is the momentum estimate of the gradient, is the second-order moment estimate of the gradient, is a constant to prevent division by zero.

10. A control system for an all-solid-state high repetition rate Q-switched laser, characterized in that: include: An all-solid-state high repetition rate Q-switched laser is configured to output current operating parameters, wherein the all-solid-state high repetition rate Q-switched laser includes a TEC module, a Q-switched module, and a wavelength adjustment module; A control module is connected to the all-solid-state high repetition rate Q-switched laser for communication, and the control module is configured to construct a target model; the target model is constructed based on the historical operating parameters of the all-solid-state high repetition rate Q-switched laser; and obtain the current operating parameters of the all-solid-state high repetition rate Q-switched laser; Preprocess the current operating parameters; input the current operating parameters into the target model, the target model outputting a first working instruction corresponding to the TEC module, a second working instruction corresponding to the Q-switching module, and a third working instruction corresponding to the wavelength adjustment module; control the TEC module to work according to the first working instruction, control the Q-switching module to work according to the second working instruction, and control the wavelength adjustment module to work according to the third working instruction.

Images