Intelligent optimization method and system for the shortest pulse in ultrafast fiber lasers

Abstract

The invention discloses an intelligent optimization method and system for the shortest pulse in an ultrafast fiber laser, and belongs to the field of automatic control of ultrafast fiber lasers. The method comprises: sampling the output signal of a mode-locked laser to form a data set; using the data set to train a recursive neural network combined with an attention mechanism, and inputting a target mode-locked spectrum to obtain an angle solution set; using a BFGS algorithm to search for an angle solution of a spectrum corresponding to the shortest pulse mode-locking, and based on this, generating a control signal to achieve automatic mode-locking; entering a spectrum monitoring mode, and if a loss of lock is detected, the maximum angle adjustment step is S max , the initial adjustment angle step is 0.2°, and the adaptive recovery algorithm is executed; if the locking mode is successfully restored, it switches to the spectrum monitoring mode, otherwise the adaptive recovery algorithm is executed; if the adjustment step is less than S max , the adaptive recovery algorithm is executed; otherwise, the network is retrained. The problem that ultrafast fiber lasers based on nonlinear polarization rotation cannot be quickly locked and are prone to losing lock is solved.

Description

Technical Field

[0001] The invention belongs to the field of automatic control of ultrafast fiber lasers, and in particular relates to an intelligent optimization method and system for the shortest pulse in an ultrafast fiber laser. Background Art

[0002] Fiber lasers have become an important tool in the field of ultrafast photonics due to their small size, low cost, good heat dissipation, and high energy conversion efficiency. They play a key role in thermal imaging, spectroscopy, communications, sensing, medical treatment, material processing, environmental monitoring, and military applications, especially in applications that require high temporal resolution and precise control. Through mode locking technology, these lasers can generate ultrashort laser pulses with a pulse width of only picoseconds or femtoseconds, which significantly improves the performance and functionality of lasers in these applications.

[0003] There are three ways to achieve mode locking: active mode locking, passive mode locking, and hybrid mode locking. Among them, passive mode locking based on nonlinear polarization rotation (NPR) has become a research hotspot due to its simple structure, narrow output pulse width, wide and flat spectrum. However, the polarization state of NPR mode-locked fiber lasers is easily affected by external disturbances, which makes mode locking control difficult. Traditional manual tuning of polarization mode locking takes a lot of time, and although some existing automatic mode locking technologies have improved efficiency, there are still problems such as long mode locking time and complex simulation systems. In particular, there are no relevant reports when searching for the shortest pulse width mode-locked pulse. Therefore, optimizing the automatic mode locking method and improving the automatic mode locking effect are important directions for current technological development. Summary of the invention

[0004] In view of the deficiencies in the prior art, the object of the present invention is to provide an intelligent optimization method for the shortest pulse in an ultrafast fiber laser, which solves the problems in the prior art.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] The smart optimization method for the shortest pulses in ultrafast fiber lasers includes the following steps:

[0007] The output spectra of the mode-locked laser at different polarization controller rotation angles are stored in the computing center via the GPIB data line to form a data set;

[0008] The dataset is used to train a recurrent neural network combined with an attention mechanism to obtain the corresponding relationship between the rotation angle of the polarization controller and the spectrum.

[0009] Input the target mode-locked spectrum into the trained recurrent neural network combined with the attention mechanism to obtain the angle solution set;

[0010] Based on the angle solution set, an adaptive optimization algorithm is used to search for the angle solution of the spectrum corresponding to the shortest pulse mode locking;

[0011] According to the search results of the adaptive optimization algorithm, a control signal is generated and fed back to the electrically controlled polarization controller to achieve automatic mode locking;

[0012] Enter the spectrum monitoring mode to monitor the spectrum output by the laser;

[0013] If the lock is lost, the maximum adjustment angle step is S max , the initial adjustment angle step is 0.2°, and the adaptive recovery algorithm is executed to recover the locked mode;

[0014] If the mode locking is successfully restored, the mode is switched to the spectrum monitoring mode, otherwise the adaptive recovery algorithm is executed;

[0015] If the adjustment step size is less than S max , an adaptive recovery algorithm is executed to recover the locked mode; otherwise, the recurrent neural network combined with the attention mechanism is retrained.

[0016] Furthermore, the recursive neural network combined with the attention mechanism includes: an input layer, an LSTM layer, an attention layer, two hidden layers and an output layer.

[0017] Furthermore, the steps of training a recurrent neural network with an attention mechanism include:

[0018] S21, divide the data set into training set and test set according to 8:2;

[0019] S22, setting loss function and optimizer;

[0020] The loss function is:

[0021]

[0022] Among them, variables a, b and c represent weighting factors, P, and I represent the peak position, predicted spectrum and target spectrum respectively, variable i represents the index of spectral data; n represents the number of spectral sampling points, and x represents the wavelength;

[0023] The optimizer uses the gradient descent algorithm, and the specific formula is as follows:

[0024]

[0025] Where α and β represent the rotation angle of the electrically controlled polarization controller, L and η represent the loss function and the learning rate that determines the descent speed, respectively;

[0026] S23, use the training set to train the model. In each training round, calculate the training loss until the loss value converges, then stop the model training;

[0027] S24, use the test set to evaluate the model.

[0028] Further, the step of using the adaptive optimization algorithm to search for an angle solution of the spectrum corresponding to the shortest pulse mode locking includes:

[0029] S41, based on the angle solution set obtained by the neural network, calculate the mode-locking spectrum corresponding to the shortest pulse that can be achieved near each angle solution, and set the initial angle solution to R = [α 0 , β 0 ], minimum step angle Δθ min ;

[0030] S42, initialize the step angle Δθ. In each iteration, the new search direction is determined by calculating the gradient, and the change direction p of Δθ is determined by the gradient. k , and update the angle to α k =α 0 +Δθ*p k ; Input the angle data into the electrically controlled polarization controller to obtain the spectrum f(α k ), if the current spectrum width is larger, then accept this adjustment and store the current spectrum for the next exploration and comparison, and change the step angle to Δθ*a, a>1; if the current spectrum width is smaller, then reject this exploration, and change the angle to Δθ*b, b<1, until the step angle is less than Δθ min , stop optimization;

[0031] S43, searching the two angles α and β one by one according to S42, and finding the maximum spectrum width solution near each angle solution;

[0032] S44, the maximum spectral width solutions of these localized parts are compared to obtain the maximum spectral width mode-locked spectrum, corresponding to the shortest pulse.

[0033] Furthermore, the spectrum monitoring mode is to continuously scan the current spectrum and compare it with any fundamental frequency mode-locked spectrum to realize the recognition of the mode-locked state, so as to monitor the loss of lock.

[0034] Furthermore, the mode-locking state identification result includes fundamental frequency mode-locking and shortest pulse width fundamental frequency mode-locking.

[0035] Furthermore, the specific steps of using the adaptive recovery algorithm to recover the locked mode are:

[0036] Obtain the current spectrum through the spectrum monitoring mode and set the maximum step angle Δθ max , the spectral difference threshold is t;

[0037] Initialize the step angle Δθ. In each iteration, the new search direction is determined by calculating the gradient, and the change direction p of Δθ is determined by the gradient. k , and update the angle to α k =α 0 +Δθ*p k ; Input the angle data into the electrically controlled polarization controller to obtain the spectrum f(α k ), if the difference between the current spectrum and the target spectrum is less than t, it means that the locking mode is restored successfully and the adjustment is stopped; if the difference between the current spectrum and the target spectrum is greater than t, it means that it is still in the unlocked state, accept this adjustment and continue the next exploration, and change the step angle to Δθ*a, a>1, until the step angle is greater than Δθ max , indicating that the recovery of mode locking failed and the exploration stopped;

[0038] If the recovery is successful, return to spectral monitoring; if the recovery fails, retrain the neural network model.

[0039] Smart optimization system for the shortest pulses in ultrafast fiber lasers, including:

[0040] Data acquisition module: stores the output spectra of the mode-locked laser at different polarization controller rotation angles to the computing center through the GPIB data line to form a data set;

[0041] Network training module: Use the data set to train a recursive neural network combined with an attention mechanism to obtain the corresponding relationship between the rotation angle of the polarization controller and the spectrum;

[0042] Angle solution acquisition module: input the target mode-locked spectrum into the trained recursive neural network combined with the attention mechanism to obtain the angle solution set;

[0043] Shortest pulse search module: Based on the angle solution set, an adaptive optimization algorithm is used to search for the angle solution of the spectrum corresponding to the shortest pulse mode locking;

[0044] Mode-locking control module: Based on the search results of the adaptive optimization algorithm, the module feeds back to the electrically controlled polarization controller to generate a control signal to achieve automatic mode-locking.

[0045] Monitoring module: Enter the spectrum monitoring mode to monitor the spectrum output by the laser;

[0046] And, locking recovery module: if the lock is detected, the maximum adjustment angle step is S max , the initial adjustment angle step is 0.2°, and the adaptive recovery algorithm is executed to restore the locking mode; if the locking mode is successfully restored, it switches to the spectrum monitoring mode, otherwise the adaptive recovery algorithm is executed; if the adjustment step is less than S max, an adaptive recovery algorithm is executed to recover the locked mode; otherwise, the recurrent neural network combined with the attention mechanism is retrained.

[0047] A computer storage medium stores a readable program, which can execute the above-mentioned intelligent optimization method for the shortest pulse in an ultrafast fiber laser when the program is running.

[0048] An electronic device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other through the communication bus;

[0049] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned intelligent optimization method for the shortest pulse in an ultrafast fiber laser.

[0050] Beneficial effects of the present invention:

[0051] The present invention utilizes an electrically controlled polarization controller, based on a recursive neural network combined with an attention mechanism and an adaptive optimization algorithm, to solve the polarization control problem in a passively mode-locked fiber laser based on an NPR structure and the search problem for the shortest pulse width mode-locked pulse, while simplifying the simulation model of the laser. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 It is a schematic diagram of the algorithm involved in the optimization method of the present invention;

[0054] Figure 2 is a neural network architecture diagram of the present invention;

[0055] Figure 3 It is a flow chart of the BFGS optimization algorithm of the present invention;

[0056] Figure 4 is a flow chart of the adaptive recovery algorithm of the present invention;

[0057] Figure 5 It is a schematic diagram of the neural network prediction result of the present invention;

[0058] Figure 6 It is a schematic diagram of the neural network training loss value result of the present invention;

[0059] Figure 7 It is a schematic diagram of the fundamental frequency mode-locked spectrum result of the present invention;

[0060] Figure 8 It is a schematic diagram of the fundamental frequency mode-locked pulse width result of the present invention;

[0061] Fig. 9 It is a schematic diagram of the fundamental frequency mode-locked spectrum result corresponding to the shortest pulse of the present invention;

[0062] Fig.10 It is a schematic diagram of the shortest pulse width result of the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] Example 1

[0065] like Figure 1 As shown in FIG. 1 , the intelligent optimization method for the shortest pulse in an ultrafast fiber laser includes four parts: a recursive neural network based on an attention mechanism, an adaptive optimization algorithm, an adaptive recovery algorithm, and spectral monitoring; the intelligent optimization method specifically includes the following steps:

[0066] S1, storing the output spectra of the mode-locked laser at different polarization controller rotation angles in the computing center via the GPIB data line to form a data set;

[0067] S2, using the data set obtained in S1 to train a neural network to obtain the corresponding relationship between the rotation angle of the polarization controller and the spectrum;

[0068] In this embodiment, the neural network is a recursive neural network combined with an attention mechanism; the recursive neural network based on the attention mechanism adds an attention weight allocation mechanism to the recursive neural network; the attention mechanism essentially assigns different calculation weights to the results of neuron outputs to achieve the function of focusing on a certain part of the numerical value; benefiting from the influence of the attention mechanism, the model removes the background noise part of the spectral data, thereby paying more attention to the data near the peak, thereby improving the running speed and accuracy of the model.

[0069] Training a recurrent neural network with attention mechanism consists of the following steps:

[0070] S21, according to the sampled data set, divide it into training set and test set, with a division ratio of 8:2;

[0071] S22, set the recursive neural network structure combined with the attention mechanism, such as Figure 2 As shown, the architecture includes: input layer, LSTM layer, attention layer, two hidden layers and output layer;

[0072] S23, setting the loss function and optimizer; the loss function is set as follows for the spectral feature:

[0073]

[0074] Among them, variables a, b and c represent weighting factors, P, and I represent the peak position, predicted spectrum and target spectrum respectively, the variable i represents the index of the spectral data; n represents the number of spectral sampling points; and x represents the wavelength.

[0075] The optimizer uses the gradient descent algorithm. The specific formula is as follows:

[0076]

[0077] Among them, α and β represent the rotation angle of the electrically controlled polarization controller, L and η represent the loss function and the learning rate that determines the descent speed, respectively.

[0078] S24, use the training set to perform model training. In each training round, the training loss is calculated until the loss value converges, then the model training is stopped.

[0079] S25, evaluate the model using the test set;

[0080] S3, input the target mode-locked spectrum into the trained neural network to obtain the angle solution set;

[0081] S4, based on the angle solution set obtained in S3, using the adaptive optimization (BFGS) algorithm, searches for the angle solution of the spectrum corresponding to the shortest pulse mode locking;

[0082] like Figure 3 As shown, the steps of searching the angular solution of the spectrum corresponding to the shortest pulse mode locking using the BFGS algorithm include:

[0083] S41, based on the angle solution set obtained by the neural network, calculate the mode-locking spectrum corresponding to the shortest pulse that can be achieved near each angle solution, and set the initial angle solution to R = [α 0 , β 0 ], minimum step angle Δθ min ;

[0084] S42, initialize the step angle Δθ. In each iteration, the new search direction is determined by calculating the gradient, and the change direction p of Δθ is determined by the gradient. k , and update the angle to α k =α 0+Δθ*p k ; Input the angle data into the electrically controlled polarization controller to obtain the spectrum f(α k ), if the current spectrum width is larger, then accept this adjustment and store the current spectrum for the next exploration and comparison, and change the step angle to Δθ*a, a>1; if the current spectrum width is smaller, then reject this exploration, and change the angle to Δθ*b, b<1, until the step angle is less than Δθ min , stop optimization;

[0085] S43, searching the two angles α and β one by one according to S42, and finding the maximum spectrum width solution near each angle solution;

[0086] S44, the maximum spectral width solutions of these localized parts are compared to obtain the maximum spectral width mode-locked spectrum, corresponding to the shortest pulse.

[0087] S5, generating an angle control signal according to the search result of the adaptive optimization algorithm and sending it to the electrically controlled polarization controller;

[0088] S6, transmitting the control signal to the electrically controlled polarization controller to achieve automatic mode locking;

[0089] S7, enter the spectrum monitoring mode to monitor the spectrum output by the laser.

[0090] The spectrum monitoring mode continuously scans the current spectrum and compares it with any fundamental frequency mode-locked spectrum to identify the mode-locked state, thereby monitoring loss of lock; the mode-locked state identification results include fundamental frequency mode-locked and minimum pulse width fundamental frequency mode-locked.

[0091] S8, if the lock is lost, the maximum adjustment angle step is S max , the initial adjustment angle step is 0.2°, and the adaptive recovery algorithm is executed to recover the locked mode;

[0092] In order to solve the problem of the sensitivity of the polarization state of light in the optical fiber to the environmental influence, the present invention proposes an adaptive recovery algorithm. The operation steps are roughly the same as those of the adaptive optimization algorithm, which can be understood as a rapid search at angles near the unlocked state. If the step length threshold is exceeded, it is judged to be unrecoverable, and the neural network will be retrained to find a new model to achieve the target pulse state.

[0093] like Figure 4 As shown in FIG. 1 , the specific steps of using the adaptive recovery algorithm to recover the locked mode are:

[0094] S81, obtain the current spectrum through the spectrum monitoring mode and set the maximum step angle Δθ max , the spectral difference threshold is t;

[0095] S82, initialize the step angle Δθ. In each iteration, the new search direction is determined by calculating the gradient, and the change direction p of Δθ is determined by the gradient. k , and update the angle to α k =α 0 +Δθ*p k ; Input the angle data into the electrically controlled polarization controller to obtain the spectrum f(α k ), if the difference between the current spectrum and the target spectrum is less than t, it means that the locking mode is restored successfully and the adjustment is stopped; if the difference between the current spectrum and the target spectrum is greater than t, it means that it is still in the unlocked state, accept this adjustment and continue the next exploration, and change the step angle to Δθ*a, a>1, until the step angle is greater than Δθ max , indicating that the recovery of mode locking failed and the exploration stopped;

[0096] S83, if the recovery is successful, return to spectrum monitoring; if the recovery fails, retrain the neural network model.

[0097] S9, if the locking mode is successfully restored, go to S7, otherwise execute the adaptive recovery algorithm;

[0098] S10, if the adjustment step is less than S max , go to S8; otherwise, go to S1, collect data again and train the neural network;

[0099] Example 2

[0100] In order to verify the effectiveness of the method proposed in Example 1, this method was used in this example to successfully predict and reproduce the single soliton mode-locked spectrum and the existing Kelly sidebands. Figure 5 As shown, it is obvious that due to the influence of the attention mechanism, the predicted spectrum is locked to the band near the central wavelength and unnecessary spectral parts are eliminated.

[0101] At the same time, due to the constraint comparison of spectral information, the method of the present invention does not misjudge the noise-like signal as a reproducible target, and the loss value calculated according to the method changes as follows Figure 6 As shown, this means that this method can accurately predict the changing relationship between the polarization controller and the spectrum and has good generalization ability.

[0102] Figure 7 and Figure 8 The fundamental frequency mode-locked spectrum and pulse width diagram are respectively achieved by the deep learning model in the experiment. The central wavelength of the spectrum is 1566.8nm, and the 3dB bandwidth is 4.6nm, indicating that the pulse width is about 0.89ps, corresponding to a pulse duration of 580fs.

[0103] Fig. 9 and Fig.10After combining with the adaptive algorithm, the shortest pulse width fundamental frequency locking was achieved in the experiment. The spectrum center wavelength is 1561.9nm, the 3dB bandwidth is 6.1nm, the pulse width is 0.71ps, and the corresponding pulse duration is 465fs.

[0104] The present invention utilizes an electrically controlled polarization controller and a recursive neural network based on an attention mechanism to realize the theoretical simulation of an ultrafast fiber laser with an NPR structure, and obtains the correspondence between the rotation angle and the spectrum. The specific training time of the model is as fast as 120.68 seconds, which greatly shortens the simulation time compared to the traditional step-by-step Fourier method, which takes several hours.

[0105] At the same time, through the target spectrum input, the corresponding mode-locking angle solution set is output to achieve automatic polarization control and fast automatic mode-locking. The fastest time to achieve automatic fundamental frequency mode-locking is only 0.18s, and the fastest time to recover from loss of lock is 0.12s. Combined with the adaptive optimization algorithm, the search for the shortest pulse width mode-locked pulse is successfully realized. The optimization of the feedback structure and the adaptive recovery mechanism enable this laser to work stably in the target state, and this idea can be applied to any spectral phenomenon.

[0106] The method of the present invention may be implemented in hardware, firmware, or as software or computer code that may be stored in a recording medium (such as a CDROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code that is originally stored in a remote recording medium or a non-temporary machine-readable medium downloaded over a network and will be stored in a local recording medium, so that the method described herein may be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that a computer, processor, microprocessor controller, or programmable hardware includes a storage component (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code, and when the software or computer code is accessed and executed by a computer, processor, or hardware, the method described herein is implemented. In addition, when a general-purpose computer accesses the code for implementing the method shown herein, the execution of the code converts the general-purpose computer into a dedicated computer for executing the method shown herein.

[0107] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. An intelligent optimization method for the shortest pulses in ultrafast fiber lasers, characterized in that: The following steps are involved: The output spectra of the mode-locked laser at different polarization controller rotation angles are stored in the computing center via the GPIB data line to form a data set; The dataset is used to train a recurrent neural network combined with an attention mechanism to obtain the corresponding relationship between the rotation angle of the polarization controller and the spectrum. Input the target mode-locked spectrum into the trained recurrent neural network combined with the attention mechanism to obtain the angle solution set; Based on the angle solution set, an adaptive optimization algorithm is used to search for the angle solution of the spectrum corresponding to the shortest pulse mode locking; According to the search results of the adaptive optimization algorithm, a control signal is generated and fed back to the electrically controlled polarization controller to achieve automatic mode locking; Enter the spectrum monitoring mode to monitor the spectrum output by the laser; If the lock is lost, the maximum adjustment angle step is S max , the initial adjustment angle step is 0.2°, and the adaptive recovery algorithm is executed to recover the locked mode; If the mode locking is successfully restored, the mode is switched to the spectrum monitoring mode, otherwise the adaptive recovery algorithm is executed; If the adjustment angle step is less than S max , then the adaptive recovery algorithm is executed to recover the locked mode; Otherwise, retrain the recurrent neural network combined with the attention mechanism; The steps of searching the angular solution of the spectrum corresponding to the shortest pulse mode locking using the adaptive optimization algorithm include: S41, based on the angle solution set obtained by the neural network, calculate the locking spectrum corresponding to the shortest pulse that can be achieved near each angle solution, set the initial angle solution to R = [α0, β0], and the minimum step angle Δθ min ; S42, initialize the step angle Δθ. In each iteration, the new search direction is determined by calculating the gradient, and the change direction p of Δθ is determined by the gradient. k , and update the angle to α k =α0+Δθ*p k ; Input the angle data into the electrically controlled polarization controller to obtain the spectrum f(α k ), if the current spectrum width is larger, then accept this adjustment and store the current spectrum for the next exploration and comparison, and change the step angle to Δθ*a, a>1; if the current spectrum width is smaller, then reject this exploration, and change the angle to Δθ*b, b<1, until the step angle is less than Δθ min , stop optimization; S43, searching the two angles α and β one by one according to S42, and finding the maximum spectrum width solution near each angle solution; S44, the maximum spectral width solutions of these localized parts are compared to obtain the maximum spectral width mode-locked spectrum, corresponding to the shortest pulse.

2. The intelligent optimization method for the shortest pulse in an ultrafast fiber laser according to claim 1, characterized in that: The recursive neural network combined with the attention mechanism includes: an input layer, an LSTM layer, an attention layer, two hidden layers and an output layer.

3. The intelligent optimization method for the shortest pulse in an ultrafast fiber laser according to claim 2, characterized in that: The steps to train a recurrent neural network with attention are: S21, divide the data set into training set and test set according to 8:2; S22, setting loss function and optimizer; The loss function is: Among them, variables a, b and c represent weighting factors, P, and I represent the peak position, predicted spectrum and target spectrum respectively, variable i represents the index of spectral data; n represents the number of spectral sampling points, and x represents the wavelength; The optimizer uses the gradient descent algorithm, and the specific formula is as follows: Among them, α and β represent the two rotation angles of the electrically controlled polarization controller, L and η represent the loss function and the learning rate that determines the descent speed, respectively; S23, use the training set to train the model. In each training round, calculate the training loss until the loss value converges, then stop the model training; S24, use the test set to evaluate the model.

4. The intelligent optimization method for the shortest pulse in an ultrafast fiber laser according to claim 1, characterized in that: The spectrum monitoring mode is to continuously scan the current spectrum and compare it with any fundamental frequency mode-locked spectrum to realize the recognition of the mode-locked state, so as to monitor the loss of lock.

5. The intelligent optimization method for the shortest pulse in an ultrafast fiber laser according to claim 4, characterized in that: The mode-locking state identification result includes fundamental frequency mode-locking and shortest pulse width fundamental frequency mode-locking.

6. The intelligent optimization method for the shortest pulse in an ultrafast fiber laser according to claim 1, characterized in that: The specific steps to recover the locked mode using the adaptive recovery algorithm are: Obtain the current spectrum through the spectrum monitoring mode and set the maximum step angle Δθ max , the spectral difference threshold is t; Initialize the step angle Δθ. In each iteration, the new search direction is determined by calculating the gradient, and the change direction p of Δθ is determined by the gradient. k , and update the angle to α k =α0+Δθ*p k ; Input the angle data into the electrically controlled polarization controller to obtain the spectrum f(α k ), if the difference between the current spectrum and the target spectrum is less than t, it means that the locking mode is restored successfully and the adjustment is stopped; if the difference between the current spectrum and the target spectrum is greater than t, it means that it is still in the unlocked state, accept this adjustment and continue the next exploration, and change the step angle to Δθ*a, a>1, until the step angle is greater than Δθ max , indicating that the recovery of mode locking failed and exploration stopped; If the recovery is successful, return to spectral monitoring; if the recovery fails, retrain the neural network model.

7. Intelligent optimization system for the shortest pulses in ultrafast fiber lasers, characterized in that include: Data acquisition module: stores the output spectra of the mode-locked laser at different polarization controller rotation angles to the computing center through the GPIB data line to form a data set; Network training module: Use the data set to train a recursive neural network combined with an attention mechanism to obtain the corresponding relationship between the rotation angle of the polarization controller and the spectrum; Angle solution acquisition module: input the target mode-locked spectrum into the trained recursive neural network combined with the attention mechanism to obtain the angle solution set; Shortest pulse search module: Based on the angle solution set, an adaptive optimization algorithm is used to search for the angle solution of the spectrum corresponding to the shortest pulse mode locking; Mode-locking control module: Based on the search results of the adaptive optimization algorithm, the module feeds back to the electrically controlled polarization controller to generate a control signal to achieve automatic mode-locking. Monitoring module: Enter the spectrum monitoring mode to monitor the spectrum output by the laser; And, locking recovery module: if the lock is detected, the maximum adjustment angle step is S max , the initial adjustment angle step is 0.2°, and the adaptive recovery algorithm is executed to recover the locked mode; If the locking mode is successfully restored, the spectrum monitoring mode is switched to, otherwise the adaptive recovery algorithm is executed; if the adjustment angle step is less than S max , then the adaptive recovery algorithm is executed to recover the locked mode; Otherwise, retrain the recurrent neural network combined with the attention mechanism; The steps of searching the angular solution of the spectrum corresponding to the shortest pulse mode locking using the adaptive optimization algorithm include: S41, based on the angle solution set obtained by the neural network, calculate the locking spectrum corresponding to the shortest pulse that can be achieved near each angle solution, set the initial angle solution to R = [α0, β0], and the minimum step angle Δθ min ; S42, initialize the step angle Δθ. In each iteration, the new search direction is determined by calculating the gradient, and the change direction p of Δθ is determined by the gradient. k , and update the angle to α k =α0+Δθ*p k ; Input the angle data into the electrically controlled polarization controller to obtain the spectrum f(α k ), if the current spectrum width is larger, then accept this adjustment and store the current spectrum for the next exploration and comparison, and change the step angle to Δθ*a, a>1; if the current spectrum width is smaller, then reject this exploration, and change the angle to Δθ*b, b<1, until the step angle is less than Δθ min , stop optimization; S43, searching the two angles α and β one by one according to S42, and finding the maximum spectrum width solution near each angle solution; S44, the maximum spectral width solutions of these localized parts are compared to obtain the maximum spectral width mode-locked spectrum, corresponding to the shortest pulse.

8. A computer storage medium storing a readable program, characterized in that: When the program is run, the intelligent optimization method for the shortest pulse in an ultrafast fiber laser as claimed in any one of claims 1 to 6 can be executed.

9. An electronic device, characterized in that: include: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the intelligent optimization method for the shortest pulse in an ultrafast fiber laser according to any one of claims 1 to 6.

Images