Design method and system of fiber pulse vortex laser based on neural network

Abstract

This invention provides a design method and system for a fiber pulsed vortex laser based on a neural network, belonging to the field of fiber pulsed vortex laser design technology. This invention measures and records the output pulse width under different structural parameters to obtain experimental sample data; it then constructs a back-prediction model of the output pulse width of the fiber pulsed vortex laser based on a neural network; next, it uses the experimental sample data to calculate the weight coefficients of the back-prediction model to determine the initial output pulse width back-prediction model; it then iteratively trains the initial output pulse width back-prediction model to obtain the final weight coefficients of the final output pulse width back-prediction model; finally, it inputs the pre-designed pulse width into the final fiber pulsed vortex laser output pulse width back-prediction model to obtain the structural parameters of the fiber pulsed vortex laser. This invention offers greater flexibility compared to existing technologies.

Description

Technical Field

[0001] This invention relates to the field of fiber pulsed vortex laser design technology, specifically to a design method and system for a fiber pulsed vortex laser based on a neural network. Background Technology

[0002] The superior propagation characteristics of vortex light in turbulent flow have attracted widespread attention. In the optical communication band, fiber pulsed vortex lasers, as a source of vortex light carrying time-domain information, often require strict control over their pulse width. Therefore, for industries with high requirements for pulse width, such as communications, laser processing, and others, there is an urgent need for a method to quickly and accurately design the width of pulsed vortex light, thereby promoting the development of the communications and laser processing industries.

[0003] Traditional methods for determining the pulse width of lasers in optical fibers involve solving Maxwell's equations using the slow-wave approximation to obtain the vector Helmholtz equation. This is then further solved in cylindrical coordinates using the scalar approximation, and the time-domain expansion is performed using perturbation theory to derive the longitudinal pulse evolution equation—the nonlinear Schrödinger equation. For picosecond-level fiber laser pulses, solving the nonlinear Schrödinger equation using the split-step Fourier transform algorithm provides a more convenient and accurate way to obtain the final evolution result of the fiber laser pulse.

[0004] However, the split-step Fourier algorithm is based on the adaptive evolution of light under Fourier transform, which is a forward calculation process. This method makes it difficult to reverse-engineer the structural parameters of the fiber pulsed vortex laser to meet the flexibility requirements such as cost savings during design when the pulse width of the fiber pulsed vortex laser is required. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a design method and system for fiber pulsed vortex lasers based on neural networks, which solves the problem that existing fiber pulsed vortex laser design methods cannot flexibly adjust structural parameters.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] In a first aspect, this invention proposes a design method for a fiber pulsed vortex laser based on a neural network, the method comprising:

[0010] S1. Measure the pulse width of the fiber pulsed vortex laser output under different structural parameters that affect the output pulse width of the fiber pulsed vortex laser, and use all the structural parameters and their corresponding pulse width experimental data as experimental sample data.

[0011] S2. Constructing a reverse prediction model for the output pulse width of a fiber pulse vortex laser based on a neural network;

[0012] S3. Based on the experimental sample data, calculate the weight coefficients of the backpropagation algorithm and the steepest descent method for the back prediction model of the output pulse width of the fiber pulse vortex laser, and determine the initial back prediction model of the output pulse width of the fiber pulse vortex laser based on the weight coefficients.

[0013] S4. Iteratively train the initial fiber pulse vortex laser output pulse width inverse prediction model until the preset conditions are met, and stop training. The weight coefficients that meet the preset conditions are used as the final weight coefficients of the final fiber pulse vortex laser output pulse width inverse prediction model.

[0014] S5. Input the pulse width of the pre-designed fiber pulsed vortex laser into the final fiber pulsed vortex laser output pulse width inverse prediction model to obtain the structural parameters of the fiber pulsed vortex laser.

[0015] Preferably, the structural parameters include easily modifiable parameters and difficult-to-modify parameters; the easily modifiable parameters include pump power, ordinary fiber length, erbium-doped fiber length, and reflectivity of broadband mirrors; the difficult-to-modify parameters include modulation depth and unsaturated loss of saturable absorbers, loss of three-port circulators, and loss of long-period fiber gratings.

[0016] Preferably, the weighting coefficients for calculating the backpropagation prediction model of the fiber pulse vortex laser output pulse width based on the experimental sample data using the backpropagation algorithm and the steepest descent method include:

[0017] The weighting coefficients for the inverse prediction model of the output pulse width of the fiber pulse vortex laser, calculated using the steepest descent method to achieve a local minimum value for the difference E between the experimental value and the model prediction value, are used as the weighting coefficients. The difference value E is expressed by the formula:

[0018]

[0019] Where ω and λ represent the weight vectors from the input structural parameter unit to the hidden layer unit and from the hidden layer unit to the output pulse width, respectively; s represents the number of experiments in the experimental sample; T s Indicates the pulse width output value in the experiment; O s This represents the pulse width output value in a neural network system.

[0020] Preferably, the preset condition is: the error between the predicted fiber pulse vortex laser pulse width and the actual width is less than a set value.

[0021] Preferably, S5 includes:

[0022] The structural parameters of the fiber pulsed vortex laser under cost optimization conditions are calculated using a nonlinear programming model; the nonlinear programming model is shown in the following equation:

[0023]

[0024]

[0025] Where, α k This represents the unit cost of the k-th input parameter; I k ω represents the normalized value of the actual measured value for the k-th parameter in the experiment; O represents the determined pulse width; jk The weights of neuronal synaptic connections from the input unit to the hidden unit; λ represents the activation function. j represents the connection weight of the next layer of neuron synapses; j represents the hidden unit number.

[0026] Secondly, this invention also proposes a design system for a fiber pulsed vortex laser based on a neural network, the system comprising:

[0027] The experimental sample data acquisition and storage module is used to measure the pulse width of the output fiber pulse vortex laser under different structural parameters, and to use all structural parameters and corresponding pulse width experimental data as experimental sample data.

[0028] A module for obtaining the inverse prediction model of the output pulse width of a fiber pulse vortex laser is used to construct an inverse prediction model of the output pulse width of a fiber pulse vortex laser based on a neural network.

[0029] The weighting coefficient calculation module is used to calculate the weighting coefficients of the fiber pulse vortex laser output pulse width inverse prediction model based on the experimental sample data using the backpropagation algorithm and the steepest descent method, and to determine the initial fiber pulse vortex laser output pulse width inverse prediction model based on the weighting coefficients.

[0030] The weight coefficient iterative learning module is used to iteratively train the initial fiber pulse vortex laser output pulse width inverse prediction model until a preset condition is met, and then stop training. The weight coefficients that meet the preset condition are used as the final weight coefficients of the final fiber pulse vortex laser output pulse width inverse prediction model.

[0031] The fiber pulsed vortex laser structural parameter acquisition module is used to input the pulse width of the pre-designed fiber pulsed vortex laser into the final fiber pulsed vortex laser output pulse width inverse prediction model to obtain the structural parameters of the fiber pulsed vortex laser.

[0032] Preferably, the structural parameters include easily modifiable parameters and difficult-to-modify parameters; the easily modifiable parameters include pump power, ordinary fiber length, erbium-doped fiber length, and reflectivity of broadband mirrors; the difficult-to-modify parameters include modulation depth and unsaturated loss of saturable absorbers, loss of three-port circulators, and loss of long-period fiber gratings.

[0033] Preferably, the weighting coefficient calculation module calculates the weighting coefficients of the fiber pulse vortex laser output pulse width inverse prediction model based on the experimental sample data using the backpropagation algorithm and the steepest descent method, including:

[0034] The weighting coefficients for the inverse prediction model of the output pulse width of the fiber pulse vortex laser, calculated using the steepest descent method to achieve a local minimum value for the difference E between the experimental value and the model prediction value, are used as the weighting coefficients. The difference value E is expressed by the formula:

[0035]

[0036] Where ω and λ represent the weight vectors from the input structural parameter unit to the hidden layer unit and from the hidden layer unit to the output pulse width, respectively; s represents the number of experiments in the experimental sample; T s Indicates the pulse width output value in the experiment; O s This represents the pulse width output value in a neural network system.

[0037] Preferably, the preset condition is: the error between the predicted fiber pulse vortex laser pulse width and the actual width is less than a set value.

[0038] Preferably, the fiber pulsed vortex laser structure parameter acquisition module, when performing the step of inputting the pre-designed pulse width of the fiber pulsed vortex laser into the final fiber pulsed vortex laser output pulse width inverse prediction model to obtain the fiber pulsed vortex laser structure parameters, includes:

[0039] The structural parameters of the fiber pulsed vortex laser under cost optimization conditions are calculated using a nonlinear programming model; the nonlinear programming model is shown in the following equation:

[0040]

[0041]

[0042] Where, α k This represents the unit cost of the k-th input parameter; I k ω represents the normalized value of the actual measured value for the k-th parameter in the experiment; O represents the determined pulse width; jk The weights of neuronal synaptic connections from the input unit to the hidden unit; λ represents the activation function.j represents the connection weight of the next layer of neuron synapses; j represents the hidden unit number.

[0043] (III) Beneficial Effects

[0044] This invention provides a design method and system for a fiber pulsed vortex laser based on a neural network. Compared with existing technologies, it has the following advantages:

[0045] 1. This invention measures and records the pulse width of the output fiber pulsed vortex laser under different structural parameters affecting its output pulse width, thereby obtaining experimental sample data. It then constructs a reverse prediction model for the output pulse width of the fiber pulsed vortex laser based on a neural network. Next, it uses the experimental sample data to calculate the weight coefficients of the reverse prediction model to determine the initial reverse prediction model. The initial reverse prediction model is iteratively trained to obtain the final weight coefficients of the final reverse prediction model. Finally, the pre-designed pulse width of the fiber pulsed vortex laser is input into the final reverse prediction model to obtain the structural parameters of the fiber pulsed vortex laser. This invention allows for the reverse design of the structural parameters of the fiber pulsed vortex laser, given a known pulse width, to meet requirements such as cost savings during design, offering greater flexibility.

[0046] 2. This invention is applicable to the pulse width design of fiber pulsed vortex lasers in various communication bands. It can reduce the repetitiveness of calculations caused by each laser design, reduce experimental errors, and provide real-time prediction of the output of a pulsed laser. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is an embodiment of a design method for a fiber pulsed vortex laser based on a neural network, as described in this invention.

[0049] Figure 2 This is a system block diagram of a design system for a neural network-based fiber pulsed vortex laser according to an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] This application provides a design method and system for fiber pulsed vortex lasers based on neural networks, which solves the problem that existing fiber pulsed vortex laser design methods cannot flexibly adjust structural parameters. It achieves the goal of flexibly designing fiber pulsed vortex lasers based on realistic conditions such as optimal cost, given clear pulse width requirements.

[0052] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0053] Example 1:

[0054] A fiber pulsed vortex laser consists of a semiconductor pump source, a 980 / 1550nm wavelength division multiplexer, erbium-doped fiber, an isolator, a saturable absorber, a three-port fiber circulator, a long-period fiber grating, a polarization controller, and a broadband mirror. Its specific working principle is as follows:

[0055] A 980nm / 1550nm wavelength division multiplexer injects a 980nm semiconductor laser into an erbium-doped fiber, causing population inversion in the erbium-doped fiber. Based on the principle of stimulated emission of laser light, population transitions to lower energy levels generate laser light in the communication band. With the help of an isolator, the laser is transmitted unidirectionally, maximizing unidirectional efficiency. Pulsed light is generated through nonlinear absorption by a saturable absorber (higher intensity, higher transmittance). A circulator guides the pulsed light into a long-period fiber grating, converting the fundamental mode pulse into a first-order mode pulse. Under the action of a polarization controller, the first-order mode pulse is converted into pulsed vortex light. A mirror reflects part of the pulsed vortex light back for oscillation, while the remaining portion is output.

[0056] Based on the structure and working principle of fiber pulsed vortex lasers, we can determine the relevant parameters affecting the output pulse width of the fiber pulsed vortex laser according to actual needs. Then, based on these parameters, we can solve for the output pulse width of the fiber pulsed vortex laser, thereby enabling the design of the fiber pulsed vortex laser. Based on this approach...

[0057] Firstly, this invention proposes a design method for a fiber pulsed vortex laser based on a neural network, see [link to relevant documentation]. Figure 1 The method includes:

[0058] S1. Measure the pulse width of the fiber pulsed vortex laser output under different structural parameters that affect the output pulse width of the fiber pulsed vortex laser, and use all the structural parameters and their corresponding pulse width experimental data as experimental sample data.

[0059] S2. Constructing a reverse prediction model for the output pulse width of a fiber pulse vortex laser based on a neural network;

[0060] S3. Based on the experimental sample data, calculate the weight coefficients of the backpropagation algorithm and the steepest descent method for the back prediction model of the output pulse width of the fiber pulse vortex laser, and determine the initial back prediction model of the output pulse width of the fiber pulse vortex laser based on the weight coefficients.

[0061] S4. Iteratively train the initial fiber pulse vortex laser output pulse width inverse prediction model until the preset conditions are met, and stop training. The weight coefficients that meet the preset conditions are used as the final weight coefficients of the final fiber pulse vortex laser output pulse width inverse prediction model.

[0062] S5. Input the pulse width of the pre-designed fiber pulsed vortex laser into the final fiber pulsed vortex laser output pulse width inverse prediction model to obtain the structural parameters of the fiber pulsed vortex laser.

[0063] As can be seen, this embodiment measures and records the pulse width of the fiber pulsed vortex laser output under different structural parameters affecting the output pulse width of the fiber pulsed vortex laser, thereby obtaining experimental sample data; and constructs a reverse prediction model for the output pulse width of the fiber pulsed vortex laser based on a neural network; then, it uses the experimental sample data to calculate the weight coefficients of the reverse prediction model to determine the initial reverse prediction model; and iteratively trains the initial reverse prediction model to obtain the final weight coefficients of the final reverse prediction model; finally, the pre-designed pulse width of the fiber pulsed vortex laser is input into the final reverse prediction model to obtain the structural parameters of the fiber pulsed vortex laser. This embodiment allows for the reverse design of the structural parameters of the fiber pulsed vortex laser to meet requirements such as cost savings during design, given a known fiber pulsed vortex laser pulse width, thus offering greater flexibility.

[0064] The following is in conjunction with the appendix Figure 1 The following details the implementation process of an embodiment of the present invention, including explanations of the specific steps S1-S5.

[0065] S1. Measure the pulse width of the fiber pulsed vortex laser output under different structural parameters that affect the output pulse width of the fiber pulsed vortex laser, and use all the structural parameters and their corresponding pulse width experimental data as experimental sample data.

[0066] 1) In this embodiment, nine structural parameters affecting the output pulse width of the fiber pulsed vortex laser are identified: pump power i1, length of ordinary fiber i2, length and concentration of erbium-doped fiber i3 and i4, modulation depth i5 and unsaturated loss i6 of the saturable absorber, loss of the three-port circulator i7, loss of the long-period fiber grating i8, and reflectivity of the broadband mirror i9. Among these, pump power, length of ordinary fiber, length of erbium-doped fiber, and reflectivity of the broadband mirror are clearly designable and modifiable for a laser; these structural parameters that are easily modified in experiments are called easily modifiable parameters. The modulation depth and unsaturated loss of the saturable absorber, the loss of the three-port circulator, and the loss of the long-period fiber grating have a certain degree of randomness and are not easily modified.

[0067] To facilitate calculation, the max-min method is first used to normalize all structural parameters:

[0068]

[0069] Where s (s=1,2,…,n) represents known experimental samples, and k (k=1,2,…,9) represents input parameters; This represents the actual measured value of the k-th parameter in the s-th experiment; i k，max This represents the maximum value of the k-th parameter across all experimental data, i.e., the upper limit of that parameter; i k，min This represents the minimum value of the k-th parameter among all experimental data, i.e., the lower limit of that parameter; This represents the normalized value of the actual measured value of the k-th parameter in the s-th experiment. Furthermore, since the experimental data needs to be normalized before calculation, and the final calculation result is also a normalized result, it is necessary to reverse-transform the final result back to the actual experimental value.

[0070] 2) Measure the pulse width of the fiber pulsed vortex laser output under different structural parameters, and use the experimental data of all the structural parameters and their corresponding pulse widths as experimental sample data.

[0071] Multiple experiments were conducted to measure the pulse width of the output fiber pulsed vortex laser under different structural parameters, and the results were saved as experimental samples. Each structural parameter and its corresponding pulse width value constituted a set of experimental sample data.

[0072] S2. Construct a reverse prediction model for the output pulse width of a fiber pulse vortex laser based on a neural network.

[0073] The reverse prediction model for the output pulse width of the fiber pulsed vortex laser constructed in this embodiment is essentially a weighted expression between the output pulse width and structural parameters. This weighted expression can be obtained by training a neural network model. In this embodiment, the neural network model uses nine different structural parameters as input units, sets three hidden layer units in the middle, and uses the pulse width of the output fiber pulsed vortex laser as the output unit. The number of hidden layer units is generally... When M is the number of output units and N is the number of input units, it is more suitable. In this embodiment, Using a neural network model, nine different structural parameters and weighted expressions for the hidden layer and the output pulse width of the fiber pulse vortex laser were established under supervised learning conditions. These weighted expressions constitute the constructed inverse prediction model for the output pulse width of the fiber pulse vortex laser.

[0074] In this embodiment, the steps for establishing the neural network model include:

[0075] (1) Normalize the actual measured value of the k-th parameter in the s-th experiment to a numerical value. As input, it is used with hidden layer units The relationship is:

[0076]

[0077]

[0078] Where, ω jk The weights represent the synaptic connections between the input unit and the hidden layer unit. A positive weight indicates that the input unit promotes the output of the first layer, while a negative weight indicates that the input unit inhibits the output of the first layer. This represents the weighted sum of the signals of each input structural parameter, which is equivalent to the signal at the end of the neuron. This represents the activation function, which acts as a non-linear mapping, converting signals at the neuron's terminals... The output size is limited to the range (0,1). The activation function is a monotonic function with upper and lower bounds of 1 and 0, respectively, and α can be used to control the slope of the region of change.

[0079] (2) Hidden layer units As the input to the next neuron, its relationship with the output pulse width of the fiber pulsed vortex laser is as follows:

[0080]

[0081] In this layer, the output of the previous neuron serves as the input of the next neuron, with no lateral information exchange. λ jThis represents the connection weights of the synapses in the next layer of neurons; a positive weight indicates activation, and a negative weight indicates inhibition. s This represents the weighted sum of the input signals of each hidden layer unit.

[0082] For any given set of structural parameters as input, the output pulse width of the fiber pulsed vortex laser is always a weighted value {λ}. j ω jk The function of}. If a suitable set of weights {λ} can be determined. j ω jk If the structural parameters of any set of experiments are a function of the output pulse width of the fiber pulsed vortex laser, then the problem of predicting the output pulse width of the fiber pulsed vortex laser is solved. This is because for any unknown fiber pulsed vortex laser, simply designing its parameters allows for accurate prediction of its output pulse width.

[0083] S3. Based on the experimental sample data, calculate the weight coefficients of the backpropagation algorithm and the steepest descent method for the output pulse width of the fiber pulse vortex laser, and determine the initial backpropagation model for the output pulse width of the fiber pulse vortex laser based on the weight coefficients.

[0084] The weighting coefficients of nine structural parameters and the hidden layers, as well as the weighting coefficients of the hidden layers and the output pulse width of the fiber pulsed vortex laser, were calculated using the backpropagation algorithm and the steepest descent method. For a multilayer neural network, if an appropriate set of weights can be calculated, the pulse width of the fiber pulsed vortex laser can be made predictable. However, in reality, the pulse width output of the fiber pulsed vortex laser in the neural network system often differs from the experimental output. Ideally, the experimental pulse width output is T. s In a neural network system, the pulse width output is O. s The difference between the two can be expressed as:

[0085]

[0086] Here, ω and λ represent the weight vectors from the input structural parameter unit to the hidden layer unit and from the hidden layer unit to the output pulse width, respectively. This formula measures the difference between the actual pulse width and the ideal pulse width output of the neural network model under a given set of weights. Therefore, it is necessary to calculate a suitable set of weights to minimize E. The activation function is a continuously differentiable nonlinear function for each weight. In this embodiment, the steepest descent method is used to calculate the local minimum of E. In the weight space composed of weight coordinates, starting from any initial position (ω0, λ0), the movement proceeds in the direction of the fastest descent (ω0, λ0), which is the negative gradient direction. Move. Only the gradient is needed. This means that E has not reached a local minimum, so it can continue to move a small distance in the weight space to reach a new position.

[0087]

[0088] μ is a parameter. As long as μ is relatively small, E(ω1, λ1) will definitely be less than the original difference value E(ω0, λ0), which means that the difference between the actual output pulse and the ideal output pulse is gradually decreasing. By continuously iterating this process, a local minimum can be obtained within the error range.

[0089] For the step from the hidden unit to the pulse width output unit, the weight adjustment for each step of the steepest descent method is as follows:

[0090]

[0091] To simplify, let

[0092]

[0093] The correction amount can be abbreviated as:

[0094]

[0095] The weights from the input structural parameter elements to the hidden elements are corrected as follows:

[0096]

[0097] Similarly, let

[0098]

[0099] The weight correction expression from the input unit to the hidden unit can then be simplified to:

[0100]

[0101] The adjustment amounts for both sets of weights above have the same expression:

[0102]

[0103] Where p represents the index of the output neuron, q represents the index of the input neuron, and V represents the magnitude of the input signal. δ represents the distance from the hidden layer unit to the output pulse width unit. s This refers to the difference between the actual pulse width output and the ideal pulse width output, as well as the weighting and correlation of hidden layer units to the pulse output unit stimulation. However, the input structural parameter units to the hidden layer units... It also additionally depends on the δ from the hidden layer unit to the output pulse width unit. sBy iterating in this way, a set of weight vectors (ω, λ) within a predetermined accuracy range is calculated. Substituting these weights into the weight expression between the output pulse width and the structural parameters, the initial fiber pulse vortex laser output pulse width inverse prediction model can be obtained.

[0104] S4. Iteratively train the initial fiber pulse vortex laser output pulse width inverse prediction model until the preset conditions are met, and stop training. The weight coefficients that meet the preset conditions are used as the final weight coefficients of the final fiber pulse vortex laser output pulse width inverse prediction model.

[0105] Iterative prediction and learning. The experiment is repeated with different input structural parameters to predict the output pulse width of the fiber vortex laser, which is then compared to the actual width measured in the experiment. If the error exceeds a set value, the experiment is used as a new sample to retrain the neural network. The input parameters, the weights of the hidden layer, and the weights of the hidden layer relative to the fiber vortex laser pulse width are updated for a new iterative training process. The learning process ends when the error between the predicted and actual fiber vortex laser pulse widths is less than the set value, resulting in an accurate inverse prediction model for the output pulse width of the fiber vortex laser.

[0106] In addition, in this embodiment, to avoid the complexity caused by unnecessary learning, a process of test prediction is adopted after each learning technique until the continuous prediction accuracy reaches a set value or above, at which point the learning is considered complete and the model is considered ready for real-time use.

[0107] S5. Input the pulse width of the pre-designed fiber pulsed vortex laser into the reverse prediction model of the output pulse width of the fiber pulsed vortex laser to obtain the structural parameters of the fiber pulsed vortex laser.

[0108] After calculation and optimization through the above S3-S4 steps, we determined the final reverse prediction model for the output pulse width of the fiber pulse vortex laser. This model clarifies the relationship between the input parameters and the pulse width. Using this model, the pulse width of the pre-designed fiber pulse vortex laser is input into the reverse prediction model for the output pulse width of the fiber pulse vortex laser, and the structural parameters of the fiber pulse vortex laser can be obtained.

[0109] Furthermore, in this embodiment, to consider both pulse width design requirements and optimal experimental cost, we use nonlinear programming to inversely calculate the structural parameters under cost-optimal conditions based on the clearly defined pulse width design requirements. Essentially, this involves adjusting easily modifiable parameters to their optimal state, and then using a normalization formula to convert the calculated easily modifiable parameter results into the actual parameters used in the experiment. The nonlinear programming model is shown in the following equation:

[0110]

[0111]

[0112] Where, α k This represents the unit cost of the k-th input parameter, and O is the predetermined pulse width. Among the input structural parameters of fiber pulsed vortex lasers, only those k = 1, 2, and 3 are easily modifiable input parameter variables. When k = 4, 5, 6, 7, 8, and 9, the parameters fluctuate around typical values, and for a selected device, their values ​​are essentially fixed.

[0113] Considering the difficulty in solving constrained nonlinear programming problems, we transform them into unconstrained minimization problems, namely:

[0114]

[0115] Can we use the steepest descent method to solve it? Specifically:

[0116] Choose any input variable [I0, u0], abbreviated as vector X0, and the modification amount in its fastest descent direction is:

[0117]

[0118] The new input vector is

[0119] x1=x0+Δx

[0120] μ is a parameter. As long as μ is relatively small, g(x1) will always be less than the original function value g(x0), which means that the cost under the constraints is continuously decreasing. By iterating continuously, a local minimum can be reached within the error range, thus obtaining the normalized input parameter I. k The final input parameters that need to be designed can be obtained by using the inverse normalization formula:

[0121] i k =I k (i k，max -i k，min )+i k，min

[0122] Once the optimal set of fiber pulsed vortex laser structural parameters is selected, the desired fiber pulsed vortex laser can be designed and manufactured by referring to these parameters.

[0123] This completes the entire process of designing a neural network-based fiber pulsed vortex laser according to this embodiment.

[0124] Example 2:

[0125] Secondly, this invention also provides a design system for a fiber pulsed vortex laser based on a neural network, see [link to relevant documentation]. Figure 2 The system includes:

[0126] The experimental sample data acquisition and storage module is used to measure the pulse width of the output fiber pulse vortex laser under different structural parameters, and to use all structural parameters and corresponding pulse width experimental data as experimental sample data.

[0127] A module for obtaining the inverse prediction model of the output pulse width of a fiber pulse vortex laser is used to construct an inverse prediction model of the output pulse width of a fiber pulse vortex laser based on a neural network.

[0128] The weighting coefficient calculation module is used to calculate the weighting coefficients of the fiber pulse vortex laser output pulse width inverse prediction model based on the experimental sample data using the backpropagation algorithm and the steepest descent method, and to determine the initial fiber pulse vortex laser output pulse width inverse prediction model based on the weighting coefficients.

[0129] The weight coefficient iterative learning module is used to iteratively train the initial fiber pulse vortex laser output pulse width inverse prediction model until a preset condition is met, and then stop training. The weight coefficients that meet the preset condition are used as the final weight coefficients of the final fiber pulse vortex laser output pulse width inverse prediction model.

[0130] The fiber pulsed vortex laser structural parameter acquisition module is used to input the pulse width of the pre-designed fiber pulsed vortex laser into the final fiber pulsed vortex laser output pulse width inverse prediction model to obtain the structural parameters of the fiber pulsed vortex laser.

[0131] Optionally, the structural parameters include easily modifiable parameters and difficult-to-modify parameters; the easily modifiable parameters include pump power, ordinary fiber length, erbium-doped fiber length, and reflectivity of broadband mirrors; the difficult-to-modify parameters include modulation depth and unsaturated loss of saturable absorbers, loss of three-port circulators, and loss of long-period fiber gratings.

[0132] Optionally, the weighting coefficient calculation module calculates the weighting coefficients of the fiber pulse vortex laser output pulse width inverse prediction model based on the experimental sample data using the backpropagation algorithm and the steepest descent method, including:

[0133] The weighting coefficients for the inverse prediction model of the output pulse width of the fiber pulse vortex laser, calculated using the steepest descent method to achieve a local minimum value for the difference E between the experimental value and the model prediction value, are used as the weighting coefficients. The difference value E is expressed by the formula:

[0134]

[0135] Where ω and λ represent the weight vectors from the input structural parameter unit to the hidden layer unit and from the hidden layer unit to the output pulse width, respectively; s represents the number of experiments in the experimental sample; T s Indicates the pulse width output value in the experiment; O s This represents the pulse width output value in a neural network system.

[0136] Optionally, the preset condition is: the error between the predicted fiber pulse vortex laser pulse width and the actual width is less than a set value.

[0137] Optionally, the fiber pulsed vortex laser structure parameter acquisition module, when performing the step of inputting the pre-designed pulse width of the fiber pulsed vortex laser into the final fiber pulsed vortex laser output pulse width inverse prediction model to obtain the fiber pulsed vortex laser structure parameters, includes:

[0138] The structural parameters of the fiber pulsed vortex laser under cost optimization conditions are calculated using a nonlinear programming model; the nonlinear programming model is shown in the following equation:

[0139]

[0140]

[0141] Where, α k This represents the unit cost of the k-th input parameter; I k ω represents the normalized value of the actual measured value for the k-th parameter in the experiment; O represents the determined pulse width; jk The weights of neuronal synaptic connections from the input unit to the hidden unit; λ represents the activation function. j represents the connection weight of the next layer of neuron synapses; j represents the hidden unit number.

[0142] It is understood that the design system for a neural network-based fiber pulsed vortex laser provided in this embodiment of the invention corresponds to the design method for a neural network-based fiber pulsed vortex laser described above. The explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding content in the design method for a neural network-based fiber pulsed vortex laser, and will not be repeated here.

[0143] In summary, compared with existing technologies, it has the following beneficial effects:

[0144] 1. This invention measures and records the pulse width of the output fiber pulsed vortex laser under different structural parameters affecting its output pulse width, thereby obtaining experimental sample data. It then constructs a reverse prediction model for the output pulse width of the fiber pulsed vortex laser based on a neural network. Next, it uses the experimental sample data to calculate the weight coefficients of the reverse prediction model to determine the initial reverse prediction model. The initial reverse prediction model is iteratively trained to obtain the final weight coefficients of the final reverse prediction model. Finally, the pre-designed pulse width of the fiber pulsed vortex laser is input into the final reverse prediction model to obtain the structural parameters of the fiber pulsed vortex laser. This invention allows for the reverse design of the structural parameters of the fiber pulsed vortex laser, given a known pulse width, to meet requirements such as cost savings during design, offering greater flexibility.

[0145] 2. This invention is applicable to the pulse width design of fiber pulsed vortex lasers in various communication bands. It can reduce the repetitiveness of calculations caused by each laser design, reduce experimental errors, and provide real-time prediction of the output of a pulsed laser.

[0146] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0147] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A design method for a fiber pulsed vortex laser based on a neural network, characterized in that, The method includes: S1. Measure the pulse width of the fiber pulsed vortex laser output under different structural parameters that affect the output pulse width of the fiber pulsed vortex laser, and use all the structural parameters and their corresponding pulse width experimental data as experimental sample data. S2. Constructing a reverse prediction model for the output pulse width of a fiber pulse vortex laser based on a neural network; S3. Based on the experimental sample data, calculate the weight coefficients of the backpropagation algorithm and the steepest descent method for the back prediction model of the output pulse width of the fiber pulse vortex laser, and determine the initial back prediction model of the output pulse width of the fiber pulse vortex laser based on the weight coefficients. S4. Iteratively train the initial fiber pulse vortex laser output pulse width inverse prediction model until the preset conditions are met, and stop training. The weight coefficients that meet the preset conditions are used as the final weight coefficients of the final fiber pulse vortex laser output pulse width inverse prediction model. S5. Input the pulse width of the pre-designed fiber pulse vortex laser into the final fiber pulse vortex laser output pulse width inverse prediction model to obtain the structural parameters of the fiber pulse vortex laser. The weighting coefficients of the backpropagation algorithm and steepest descent method used to calculate the output pulse width inverse prediction model of the fiber pulse vortex laser based on the experimental sample data include: The steepest descent method is used to calculate the difference between the experimental value and the model prediction. The weighting coefficients at the local minimum are used as the weighting coefficients of the fiber pulse vortex laser output pulse width inverse prediction model; where the difference value... Expressed as a formula: in, These represent the weight vectors from the input structural parameter unit to the hidden layer unit and from the hidden layer unit to the output pulse width, respectively. Indicates the number of experiments in the experimental sample; This represents the pulse width output value in the experiment; This represents the pulse width output value in a neural network system. S5 includes: The structural parameters of the fiber pulsed vortex laser under cost optimization conditions are calculated using a nonlinear programming model; the nonlinear programming model is shown in the following equation: in, It indicates the first Unit cost of each input parameter; Indicates the first experiment The normalized value of the actual measured value for each parameter; This indicates the predetermined pulse width; The weights of neuronal synaptic connections from the input unit to the hidden unit; Represents the activation function; This represents the connection weights of the synapses of the next layer of neurons. This indicates the hidden unit number.

2. The method as described in claim 1, characterized in that, The structural parameters include easily modifiable parameters and difficult-to-modify parameters; the easily modifiable parameters include pump power, ordinary fiber length, erbium-doped fiber length, and reflectivity of broadband mirrors; the difficult-to-modify parameters include modulation depth and unsaturated loss of saturable absorbers, loss of three-port circulators, and loss of long-period fiber gratings.

3. The method as described in claim 1, characterized in that, The preset condition is that the error between the predicted fiber pulse vortex laser pulse width and the actual width is less than a set value.

4. A design system for a fiber pulsed vortex laser based on a neural network, characterized in that, The system includes: The experimental sample data acquisition and storage module is used to measure the pulse width of the output fiber pulse vortex laser under different structural parameters, and to use all structural parameters and corresponding pulse width experimental data as experimental sample data. A module for obtaining the inverse prediction model of the output pulse width of a fiber pulse vortex laser is used to construct an inverse prediction model of the output pulse width of a fiber pulse vortex laser based on a neural network. The weighting coefficient calculation module is used to calculate the weighting coefficients of the fiber pulse vortex laser output pulse width inverse prediction model based on the experimental sample data using the backpropagation algorithm and the steepest descent method, and to determine the initial fiber pulse vortex laser output pulse width inverse prediction model based on the weighting coefficients. The weight coefficient iterative learning module is used to iteratively train the initial fiber pulse vortex laser output pulse width inverse prediction model until a preset condition is met, and then stop training. The weight coefficients that meet the preset condition are used as the final weight coefficients of the final fiber pulse vortex laser output pulse width inverse prediction model. The fiber pulse vortex laser structure parameter acquisition module is used to input the pulse width of the pre-designed fiber pulse vortex laser into the final fiber pulse vortex laser output pulse width inverse prediction model to obtain the fiber pulse vortex laser structure parameters. The weighting coefficient calculation module calculates the weighting coefficients of the fiber pulse vortex laser output pulse width inverse prediction model based on the experimental sample data using the backpropagation algorithm and the steepest descent method. The steepest descent method is used to calculate the difference between the experimental value and the model prediction. The weighting coefficients at the local minimum are used as the weighting coefficients of the fiber pulse vortex laser output pulse width inverse prediction model; where the difference value... Expressed as a formula: in, These represent the weight vectors from the input structural parameter unit to the hidden layer unit and from the hidden layer unit to the output pulse width, respectively. Indicates the number of experiments in the experimental sample; This represents the pulse width output value in the experiment; This represents the pulse width output value in a neural network system. The fiber pulsed vortex laser structure parameter acquisition module performs the following steps when it inputs the pre-designed pulse width of the fiber pulsed vortex laser into the final fiber pulsed vortex laser output pulse width inverse prediction model to obtain the fiber pulsed vortex laser structure parameters: The structural parameters of the fiber pulsed vortex laser under cost optimization conditions are calculated using a nonlinear programming model; the nonlinear programming model is shown in the following equation: in, It indicates the first Unit cost of each input parameter; Indicates the first experiment The normalized value of the actual measured value for each parameter; This indicates the predetermined pulse width; The weights of neuronal synaptic connections from the input unit to the hidden unit; Represents the activation function; This represents the connection weights of the synapses of the next layer of neurons. This indicates the hidden unit number.

5. The system as described in claim 4, characterized in that, The structural parameters include easily modifiable parameters and difficult-to-modify parameters; the easily modifiable parameters include pump power, ordinary fiber length, erbium-doped fiber length, and reflectivity of broadband mirrors; the difficult-to-modify parameters include modulation depth and unsaturated loss of saturable absorbers, loss of three-port circulators, and loss of long-period fiber gratings.

6. The system as described in claim 4, characterized in that, The preset condition is that the error between the predicted fiber pulse vortex laser pulse width and the actual width is less than a set value.

Images