A center wavelength tunable intelligent mode-locked fiber laser and an intelligent mode-locked method

Abstract

The application discloses a center wavelength tunable intelligent mode-locked fiber laser and an intelligent mode-locked method, and relates to the technical field of lasers.The laser comprises a pump source, a wavelength division multiplexer, a gain optical fiber, an intelligent mode-locked effect component, an output component, a spectrum analyzer and a computer terminal which are sequentially connected.The output component is further connected with a transmission end of the wavelength division multiplexer.The intelligent mode-locked effect component comprises a first electronic polarization controller, a polarization-dependent isolator, a polarization maintaining optical fiber and a second electronic polarization controller which are sequentially connected.The computer terminal is used for gradually adjusting control voltages of the first electronic polarization controller and the second electronic polarization controller according to a center wavelength and a bandwidth of a spectrum output by the spectrum analyzer by using a reinforcement learning method, so that mode locking is realized at a target center wavelength.The application realizes automatic mode locking of the mode-locked fiber laser at a required center wavelength by the intelligent mode-locked method.

Description

Technical Field

[0001] This application relates to the field of laser technology, and in particular to a smart mode-locked fiber laser with tunable center wavelength and a smart mode-locking method. Background Technology

[0002] Mode-locked fiber lasers are widely used in scientific, industrial, and medical fields due to their ability to generate stable ultrashort pulses. Traditional mode-locking control methods rely on manual adjustment or simple feedback control strategies, which are often inefficient and difficult to adapt to complex and changing operating environments. With the development of intelligent control technology, intelligent mode-locking algorithms based on deep learning and reinforcement learning have emerged. These algorithms can automatically adjust the laser's operating state to adapt to different application requirements. Existing intelligent mode-locked lasers mainly focus on the search and stable maintenance of the mode-locked state, while their dynamic tuning capability for the center wavelength is limited. Therefore, there is a need for a laser that can achieve automatic mode-locking at the target center wavelength using reinforcement learning methods. Summary of the Invention

[0003] The purpose of this application is to provide a smart mode-locked fiber laser with a tunable center wavelength and a smart mode-locking method, which can achieve automatic mode-locking at the target center wavelength.

[0004] To achieve the above objectives, this application provides the following solution:

[0005] In a first aspect, this application provides a smart mode-locked fiber laser with a tunable center wavelength, comprising:

[0006] The pump source, wavelength division multiplexer, gain fiber, smart mode-locking effect component, output component, spectrometer, and computer are connected in sequence; the output component is also connected to the transmission end of the wavelength division multiplexer.

[0007] The intelligent mode-locking effect component includes: a first electronic polarization controller, a polarization-dependent isolator, a polarization-maintaining fiber, and a second electronic polarization controller connected in sequence; the computer terminal is used to gradually adjust the control voltage of the first electronic polarization controller and the second electronic polarization controller according to the center wavelength and bandwidth of the spectrum output by the spectrometer using reinforcement learning methods, so as to achieve mode-locking at the target center wavelength.

[0008] In one embodiment, the output component includes a first output coupler, a second output coupler, and a jumper connector; the input terminal of the first output coupler is connected to a second electronic polarization controller; the first output terminal of the first output coupler is connected to the transmission terminal of a wavelength division multiplexer; the second output terminal of the first output coupler is connected to the input terminal of the second output coupler; both output terminals of the second output coupler are connected to the jumper connector; one output terminal of the jumper connector serves as the output terminal of a smart mode-locked fiber laser, and the other output terminal of the jumper connector is connected to a spectrometer.

[0009] In one embodiment, the gain fiber is an erbium-doped fiber with an excitation wavelength of 1530nm-1580nm.

[0010] In one embodiment, the gain fiber is a ytterbium-doped fiber with an excitation wavelength of 1030 nm-1064 nm.

[0011] In one embodiment, the first output coupler is a 9:1 output coupler; the first output terminal is a 90% output terminal; and the second output terminal is a 10% output terminal.

[0012] In one embodiment, the second output coupler is a 5:5 output coupler.

[0013] In one embodiment, the operating wavelengths of the wavelength division multiplexer, the first output coupler, the second output coupler, and the polarization-dependent isolator are all consistent with the excitation wavelength of the gain fiber.

[0014] In one embodiment, the jumper connector is an APC output jumper connector.

[0015] Secondly, this application provides a center wavelength tunable intelligent mode-locking method, which is applied to the center wavelength tunable intelligent mode-locked fiber laser, and the intelligent mode-locking method includes:

[0016] Acquire the output spectrum of the spectrometer and extract the center wavelength and bandwidth of the output spectrum;

[0017] Based on the center wavelength and bandwidth, the reinforcement learning method is used to dynamically adjust the control voltages of the first and second electronic polarization controllers to achieve stable mode-locking at the desired center wavelength.

[0018] In one embodiment, the training process of the reinforcement learning method includes multiple training stages, and each training stage includes multiple training rounds;

[0019] In each training round, a training control voltage for each training round is randomly generated as an initial voltage parameter; the training control voltage includes a first training control voltage and a second training control voltage.

[0020] According to the strategy of the reinforcement learning method, actions are progressively selected to act on the initial voltage parameters to gradually change the initial voltage parameters, thereby obtaining the center wavelength and bandwidth of the output spectrum after each change of voltage parameters; the actions include incremental or decremental operations on the training control voltage.

[0021] Based on the center wavelength and bandwidth of the changing output spectrum, the reward value corresponding to each training round is calculated using a reward function; the changing output spectrum includes the output spectrum before each change of voltage parameters and the output spectrum after each change of voltage parameters.

[0022] The policy of the reinforcement learning method is updated using the soft actor-critic algorithm based on the reward value;

[0023] Repeat the training rounds until the strategy converges to the set threshold.

[0024] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0025] This application provides a smart mode-locked fiber laser with tunable center wavelength and a smart mode-locking method. A smart mode-locking effect component, a spectrometer, and a controller are incorporated into a conventional laser. The smart mode-locking effect component includes a first electronic polarization controller, a polarization-dependent isolator, a polarization-maintaining fiber, and a second electronic polarization controller connected in sequence. A computer uses reinforcement learning to adjust the control voltages of the first and second electronic polarization controllers of the smart mode-locking effect component. By adjusting the control voltages of the two voltage polarization controllers, the output spectrum is changed, achieving automatic mode-locking. The first electronic polarization controller, the polarization-dependent isolator, and the second electronic polarization controller constitute a nonlinear polarization rotation structure to provide the mode-locking effect. The polarization-dependent isolator, the polarization-maintaining fiber, and the second electronic polarization controller constitute a Lyot filter, allowing the center wavelength of the mode-locked light to be tuned. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1This is a schematic diagram of a smart mode-locked fiber laser with tunable center wavelength according to one embodiment of this application.

[0028] Figure labels: 1-Pump source, 2-Wavelength division multiplexer, 3-Gain fiber, 4-First electronic polarization controller, 5-Polarization correlation isolator, 6-Polarization maintaining fiber, 7-Second electronic polarization controller, 8-9:1 output coupler, 9-5:5 output coupler, 10-APC output jumper, 11-Spectrometer, 12-Computer terminal. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] Existing intelligent mode-locked lasers mainly focus on the search and stable maintenance of mode-locked states, while their ability to dynamically tune the center wavelength is limited. Many applications, such as spectral analysis, optical communication, and biomedical imaging, have stringent requirements for the center wavelength of the laser pulse, necessitating an intelligent mode-locked laser capable of real-time and precise adjustment of the center wavelength. This application provides an intelligent mode-locked fiber laser with a tunable center wavelength, which not only achieves automatic mode-locking but also possesses the ability to automatically mode-lock at any center wavelength within the tuning range, thus improving the stability of the mode-locked fiber laser.

[0032] like Figure 1 As shown, this application provides a smart mode-locked fiber laser with tunable center wavelength, comprising: a pump source 1, a wavelength division multiplexer 2, a gain fiber 3, a smart mode-locking effect component, an output component, a spectrum analyzer 11, and a computer terminal 12 connected in sequence; the output component is also connected to the transmission end of the wavelength division multiplexer 2; the smart mode-locking effect component comprises: a first electronic polarization controller 4, a polarization correlation isolator 5, a polarization-maintaining fiber 6, and a second electronic polarization controller 7 connected in sequence; the computer terminal is used to analyze the center wavelength and bandwidth of the output spectrum of the spectrum analyzer 11 using reinforcement learning methods to gradually adjust the control voltages of the first electronic polarization controller 4 and the second electronic polarization controller 7 to achieve mode-locking at the target center wavelength.

[0033] The laser described above can automatically adjust the voltage of the electronic polarization controller, thereby precisely controlling the polarization state of the light field within the laser cavity to achieve mode-locking at the target center wavelength. Compared with existing technologies, the laser of this application not only automatically achieves mode-locking but also has the ability to automatically perform mode-locking at any center wavelength within the tuning range, significantly improving the flexibility and application range of the laser.

[0034] In one exemplary embodiment, the output component includes a first output coupler, a second output coupler, and two jumper connectors; the input terminal of the first output coupler is connected to the second electronic polarization controller 7; the first output terminal of the first output coupler is connected to the transmission terminal of the wavelength division multiplexer 2; the second output terminal of the first output coupler is connected to the input terminal of the second output coupler; both output terminals of the second output coupler are connected to the jumper connectors; the output terminal of one jumper connector serves as the output terminal of the intelligent mode-locked fiber laser, and the output terminal of the other jumper connector is connected to the spectrometer 11.

[0035] In practical applications, the first output coupler is a 9:1 output coupler 8; the first output terminal is a 90% output terminal; and the second output terminal is a 10% output terminal. The second output coupler is a 5:5 output coupler 9.

[0036] In this application, the ring cavity assembly includes: a standard single-mode fiber, a gain fiber 3, a smart mode-locking effect component, a wavelength division multiplexer 2, and a 9:1 output coupler 8. The pump source 1 is connected to the ring cavity assembly. The ring cavity devices are connected by single-mode fibers transmitting in the corresponding wavelength band.

[0037] In one exemplary embodiment, the gain fiber 3 is an erbium-doped fiber with an excitation wavelength of 1530nm-1580nm.

[0038] In another exemplary embodiment, the gain fiber 3 is a ytterbium-doped fiber with an excitation wavelength of 1030nm-1064nm.

[0039] In one exemplary embodiment, the operating wavelengths of the wavelength division multiplexer 2, the first output coupler, the second output coupler, and the polarization-dependent isolator 5 are all consistent with the excitation wavelength of the gain fiber 3.

[0040] In one exemplary embodiment, the jumper connector is an APC output jumper 10, which is an Angled Physical Contact (APC) output jumper.

[0041] In an exemplary embodiment, this application also provides specific connections for various devices in a center wavelength tunable smart mode-locked fiber laser. The reflecting end of wavelength division multiplexer 2 is connected to pump source 1; the common end of wavelength division multiplexer 2 is connected to one end of gain fiber 3; the other end of gain fiber 3 is connected to one end of smart mode-locking effect component; smart mode-locking effect component is used to provide saturable absorption effect and filtering effect; the other end of smart mode-locking effect component is connected to the input end of 9:1 output coupler 8; 90% output end of 9:1 output coupler 8 is connected to the transmission end of wavelength division multiplexer 2; 10% output end of 9:1 output coupler 8, as the output end of ring cavity, is connected to the input end of 5:5 output coupler 9; the other two ends of 5:5 output coupler 9 are both connected to APC output jumper 10; one of them is optionally connected to spectrum analyzer 11, and spectrum analyzer 11 is connected to computer terminal 12 via network interface; the output light of the other APC output jumper 10 will be used as the final output light; the soft in computer terminal 12 The actor-critic reinforcement learning method enables lasers to output stable, intelligently tunable mode-locked light.

[0042] The intelligent mode-locking effect component includes: a first electronic polarization controller 4, a polarization-dependent isolator 5, a polarization-maintaining fiber 6, and a second electronic polarization controller 7. The polarization-dependent isolator 5 is connected to the other end of the gain fiber 3. The first electronic polarization controller 4 is positioned between the polarization-dependent isolator 5 and the gain fiber 3. The other end of the polarization-dependent isolator 5 is connected to one end of the polarization-maintaining fiber 6. The other end of the polarization-maintaining fiber 6 is connected to a 9:1 output coupler 8. The second electronic polarization controller 7 is positioned between the polarization-maintaining fiber 6 and the 9:1 output coupler 8. The first electronic polarization controller 4 and the second electronic polarization controller 7 are connected to a computer terminal 12 via a network interface.

[0043] In practical applications, the computer terminal 12 uses the soft actor-critic reinforcement learning method to analyze spectral information. Based on the required center wavelength, it adjusts the polarization state of the light in the cavity by changing the control voltage of the first electronic polarization controller 4 and the second electronic polarization controller 7 to achieve stable mode locking at the required center wavelength.

[0044] The tuning principle of the intelligent mode-locked laser is as follows: the first electronic polarization controller 4, the polarization-dependent isolator 5, and the second electronic polarization controller 7 form a nonlinear polarization rotation structure to provide the mode-locking effect; in addition, the polarization-dependent isolator 5, the polarization-maintaining fiber 6, and the second electronic polarization controller 7 form a Lyot filter, which enables the center wavelength of the mode-locked light to be tuned; the soft actor-critic deep reinforcement learning method deployed in the computer terminal 12 can analyze the spectral information collected by the spectral analyzer 11, and adjust the polarization state of the light in the cavity by changing the control voltage of the first electronic polarization controller 4 and the second electronic polarization controller 7 according to the required center wavelength, so as to achieve stable mode-locking at the required center wavelength.

[0045] This application provides a smart mode-locked fiber laser with a tunable center wavelength. Its technological advantage lies in utilizing reinforcement learning technology to achieve adaptive control of the laser system. Specifically, through interaction between the intelligent agent and the laser system, the laser of this application can automatically adjust the voltage of the electronic polarization controller, thereby finely controlling the polarization state of the light field within the laser cavity to achieve mode-locking at the target center wavelength. Compared with existing technologies, the laser of this application not only automatically performs mode-locking operations but also possesses the ability to automatically perform mode-locking at any center wavelength within the tuning range, significantly improving the laser's flexibility and application range. This intelligent control mechanism enables the laser to quickly respond to different operating conditions, achieving efficient and stable mode-locked output while maintaining ease of operation and system robustness.

[0046] In another exemplary embodiment, a center wavelength tunable intelligent mode-locking method is provided. This method is applied to the aforementioned center wavelength tunable intelligent mode-locked fiber laser. The intelligent mode-locking method includes: acquiring the output spectrum of a spectral analyzer and extracting the center wavelength and bandwidth of the output spectrum; and dynamically adjusting the control voltages of a first electronic polarization controller and a second electronic polarization controller using a reinforcement learning method based on the center wavelength and bandwidth to achieve stable mode-locking at the desired center wavelength. Specifically, the softactor-critic deep reinforcement learning method is used to analyze spectral information and extract the center wavelength and bandwidth of the output spectrum. Based on the desired center wavelength, the softactor-critic agent continuously changes the control voltages of the first and second electronic polarization controllers to adjust the polarization state of the light within the cavity, thereby achieving stable mode-locking at the desired center wavelength.

[0047] In an exemplary embodiment, the training process of the reinforcement learning method includes multiple training stages, each training stage including multiple training rounds. In each training round, a training control voltage is randomly generated as an initial voltage parameter. The training control voltage includes a first training control voltage and a second training control voltage. Actions are progressively selected to act on the initial voltage parameter according to the policy of the reinforcement learning method, so as to progressively change the initial voltage parameter and obtain the center wavelength and bandwidth of the output spectrum after each change of the voltage parameter. The actions include incrementing or decrementing the training control voltage. Based on the center wavelength and bandwidth of the changed output spectrum, a reward value corresponding to the training round is calculated through a reward function. The changed output spectrum includes the output spectrum before each change of the voltage parameter and the output spectrum after each change of the voltage parameter. The policy of the reinforcement learning method is updated according to the reward value using a soft actor-critic algorithm. The training rounds are repeated until the policy converges to a set threshold.

[0048] Agent, environment, policy, and reward are all concepts related to reinforcement learning. The agent refers to the reinforcement learning algorithm module running on a PC, specifically the implementation of the soft actor-critic algorithm, used to select actions based on the environmental state. The environment refers to the physical system comprised of the intelligent mode-locked fiber laser system and its related components. The state of the environment is represented by spectral data (such as center wavelength and bandwidth) output by a spectral analyzer. When the agent performs an action (such as adjusting the control voltage), the environment changes accordingly, generating a new state (i.e., new spectral data). The feedback from the environment includes the new spectral data and a reward value calculated according to the reward function. The policy refers to the rule or function by which the agent selects actions based on the current environmental state, defined and optimized by the soft actor-critic algorithm, with the goal of maximizing the cumulative reward value.

[0049] The corresponding reward value is determined based on the center wavelength and bandwidth of the spectrum before and after the change, and the specific expression is as follows:

[0050] R(λ t ,Δλ t )=-w1|λ t -λ target | 2 +w2R(Δλ t )

[0051]

[0052] Where w1 and w2 are scaling factors, λ target λ is the target center wavelength. t Δλ is the center wavelength of the current step spectrum. tLet λ be the 3dB bandwidth of the current step spectrum, ReLU(·) be a commonly used activation function in deep learning, R(·) be the bandwidth reward value, and Δλ be the bandwidth reward value. G For Δλ t The threshold, when Δλ t <Δλ G The laser system is considered to be operating freely, when Δλ t ≥Δλ G At this time, the laser is considered to be in a mode-locked state, α is a discount factor used to reduce the influence of future rewards, T is the total number of search steps in each search cycle, and α(Tt) is used to minimize the number of search steps in a search cycle. w1, w2, α, Δλ G is a hyperparameter, and t is the current step.

[0053] In practical applications, reinforcement learning methods trained at one central wavelength can be fine-tuned through transfer learning to be applied to other central wavelengths, thereby reducing training time.

[0054] Transfer learning steps include:

[0055] When the target center wavelength changes, the target wavelength parameter in the reward function is updated according to the changed target center wavelength.

[0056] Save and reuse the reinforcement learning strategy that has been trained at any central wavelength.

[0057] Based on the updated reward function and the continued strategy, further training is performed on the changed target center wavelength.

[0058] By using the reinforcement learning strategy that has already been trained, the number of training rounds for the new target center wavelength is reduced, thereby reducing training time.

[0059] This application not only enables automatic mode-locking, but also has the ability to perform automatic mode-locking at any center wavelength within the tuning range, thereby improving the stability of mode-locked fiber lasers.

[0060] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0061] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A smart mode-locked fiber laser with tunable center wavelength, characterized in that, The center wavelength tunable intelligent mode-locked fiber laser includes: a pump source, a wavelength division multiplexer, a gain fiber, an intelligent mode-locking effect component, an output component, a spectrum analyzer, and a computer terminal connected in sequence; the output component is also connected to the transmission end of the wavelength division multiplexer. The intelligent mode-locking effect component includes: a first electronic polarization controller, a polarization-dependent isolator, a polarization-maintaining fiber, and a second electronic polarization controller connected in sequence; the computer terminal is used to gradually adjust the control voltage of the first electronic polarization controller and the second electronic polarization controller according to the center wavelength and bandwidth of the spectrum output by the spectrometer using reinforcement learning methods, so as to achieve mode-locking at the target center wavelength; The tuning principle of the intelligent mode-locked laser is as follows: a nonlinear polarization rotation structure is formed by a first electronic polarization controller, a polarization-dependent isolator, and a second electronic polarization controller to provide the mode-locking effect; in addition, the polarization-dependent isolator, polarization-maintaining fiber, and the second electronic polarization controller form a Lyot filter, which allows the center wavelength of the mode-locked light to be tuned; an intelligent agent deployed on the computer can analyze the spectral information collected by the spectrometer, and adjust the polarization state of the light in the cavity by changing the control voltage of the first and second electronic polarization controllers according to the required center wavelength, so as to achieve stable mode-locking at the required center wavelength; The agent refers to the reinforcement learning algorithm module running on a PC, specifically the implementation of the soft actor-critic algorithm, used to select actions based on the environmental state. The environment refers to the physical system composed of the intelligent mode-locked fiber laser system and its related components, and the state of the environment is represented by spectral data output by a spectral analyzer. When the agent performs actions, the environment changes according to these actions and generates new states. The feedback from the environment includes new spectral data and a reward value calculated according to the reward function. The policy refers to the rule or function by which the agent selects actions based on the current environmental state, defined and optimized by the soft actor-critic algorithm, with the goal of maximizing the cumulative reward value. The corresponding reward value is determined based on the center wavelength and bandwidth of the spectrum before and after the change, and the specific expression is as follows: in, , As a scaling factor, For the target center wavelength, The center wavelength of the current step spectrum. The 3dB bandwidth of the current step spectrum. This is a commonly used activation function in deep learning. It is the bandwidth bonus value. for The threshold, when The laser system is considered to operate freely, when At this time, the laser is considered to be in a mode-locked state. It's a discount factor used to reduce the impact of future rewards. It is the total number of search steps in each search cycle. It is used to minimize the number of search steps in a search cycle. , , , is a hyperparameter, and t is the current step.

2. The intelligent mode-locked fiber laser with tunable center wavelength according to claim 1, characterized in that, The output component includes a first output coupler, a second output coupler, and two jumper connectors; the input terminal of the first output coupler is connected to a second electronic polarization controller; the first output terminal of the first output coupler is connected to the transmission terminal of a wavelength division multiplexer; the second output terminal of the first output coupler is connected to the input terminal of the second output coupler; both output terminals of the second output coupler are connected to the jumper connectors; one output terminal of the jumper connector serves as the output terminal of a smart mode-locked fiber laser, and the other output terminal of the jumper connector is connected to a spectrometer.

3. The intelligent mode-locked fiber laser with tunable center wavelength according to claim 1, characterized in that, The gain fiber is an erbium-doped fiber with an excitation wavelength of 1530nm-1580nm.

4. The intelligent mode-locked fiber laser with tunable center wavelength according to claim 1, characterized in that, The gain fiber is a ytterbium-doped fiber with an excitation wavelength of 1030nm-1064nm.

5. The intelligent mode-locked fiber laser with tunable center wavelength according to claim 2, characterized in that, The first output coupler is a 9:1 output coupler; the first output terminal is a 90% output terminal; and the second output terminal is a 10% output terminal.

6. The intelligent mode-locked fiber laser with tunable center wavelength according to claim 2, characterized in that, The second output coupler is a 5:5 output coupler.

7. The intelligent mode-locked fiber laser with tunable center wavelength according to claim 2, characterized in that, The operating wavelengths of the wavelength division multiplexer, the first output coupler, the second output coupler, and the polarization-dependent isolator are all consistent with the excitation wavelength of the gain fiber.

8. The intelligent mode-locked fiber laser with tunable center wavelength according to claim 2, characterized in that, The jumper connector is an APC output jumper connector.

9. A smart mode-locking method with tunable center wavelength, characterized in that, The center wavelength tunable intelligent mode-locking method is applied to the center wavelength tunable intelligent mode-locked fiber laser according to any one of claims 1-8, wherein the intelligent mode-locking method includes: Acquire the output spectrum of the spectrometer and extract the center wavelength and bandwidth of the output spectrum; Based on the center wavelength and bandwidth, the reinforcement learning method is used to dynamically adjust the control voltages of the first and second electronic polarization controllers to achieve stable mode-locking at the desired center wavelength.

10. The intelligent mode-locking method with tunable center wavelength according to claim 9, characterized in that, The training process of the reinforcement learning method includes multiple training stages, and each training stage includes multiple training rounds. In each training round, a training control voltage is randomly generated for each training round as an initial voltage parameter; the training control voltage includes a first training control voltage and a second training control voltage. According to the strategy of the reinforcement learning method, actions are progressively selected to act on the initial voltage parameters to gradually change the initial voltage parameters, thereby obtaining the center wavelength and bandwidth of the output spectrum after each change in voltage parameters; the actions include incremental or decremental operations on the training control voltage; Based on the center wavelength and bandwidth of the changing output spectrum, the reward value corresponding to each training round is calculated using a reward function; the changing output spectrum includes the output spectrum before each change of voltage parameters and the output spectrum after each change of voltage parameters. The policy of the reinforcement learning method is updated using the soft actor-critic algorithm based on the reward value; Repeat the training rounds until the strategy converges to the set threshold.

Images