Grating design method and system for an optical coupler

Abstract

The application discloses a grating design method and system of an optical coupler, and belongs to the technical field of AR optical display. The method comprises the following steps: obtaining a noise vector to be modulated based on a working wavelength of an optical coupler to be designed; and obtaining a target grating based on the noise vector to be modulated, a pre-trained grating mapping model and a pre-trained grating optimization model. The training process of the pre-trained grating mapping model comprises the following steps: obtaining a plurality of historical noise vectors based on the optical coupler at a plurality of preset working wavelengths; updating a preset initial mapping model based on the plurality of historical noise vectors; obtaining a mapping training index and a training mapping model; if a preset condition is not met, updating the training mapping model and re-executing the training process until a preset training end condition is met, and taking the training mapping model as the pre-trained grating mapping. The grating design method of the optical coupler can perform multi-angle flat-band joint optimization on the grating of the optical coupler.

Description

Technical Field

[0001] This invention relates to the field of AR optical display technology, and in particular to a grating design method and system for an optical coupler. Background Technology

[0002] Currently, most AR devices consist of microdisplays and optical couplers (AR combiners), with the optical coupler being the key component for achieving virtual-real fusion. To achieve high-fidelity overlay of virtual and real scenes, the optical coupler must simultaneously meet two key performance requirements: first, low-loss transmission of light from the real scene to ensure clear visibility of the real world; and second, precise wavefront manipulation and directional deflection of light from the virtual image to efficiently couple it into the human eye, achieving augmented reality effects. However, traditional free-space optical couplers are too large to meet the miniaturization requirements of AR devices. In recent years, grating devices based on narrowband nonlocal metasurfaces have become a research hotspot in this field due to their unique modes such as bound states in the continuous (BIC) and guided-mode resonance (GMR). They possess natural narrowband spectral selectivity, providing a new path to resolve the contradiction between the size and performance of optical couplers.

[0003] Existing narrowband wavefront modulation methods are mainly divided into two types: The first is narrowband resonance modulation based on BIC (Browser Induction Channel). This method achieves extremely high Q values ​​and ultra-narrow linewidth resonance by strictly suppressing the radiation channel and forming an ideal BIC or quasi-BIC under specific momentum and symmetry conditions. Theoretically, it can achieve near-complete transparency in non-resonant bands, thereby reducing scene light loss. However, the excitation conditions of the BIC mode are highly angle-dependent, and it can only work effectively at specific incident angles. Once deviating from this angle, the resonance effect disappears rapidly, making it impossible to effectively deflect virtual light within a wide field of view, thus limiting the field of view expansion of AR devices. The second type is narrowband guided mode coupling modulation based on GMR (Glass Momentum Matching). It relies on the Bragg momentum matching of free-space light by a periodic structure to couple the incident light into a quasi-guided mode in the waveguide. This guided mode repeatedly leaks between the waveguide and the modulation structure and undergoes coherent interference, forming a high-Q resonance only under specific wavelength and angle conditions, thus naturally generating a narrowband spectral response. However, the resonant wavelength of the GMR mode shifts drastically with the incident angle. This strong dispersion characteristic makes it impossible to stably and accurately deflect virtual image light towards the user's eyes across a wide field of view. It is only suitable for eye-tracking scenarios where the angle is not sensitive, and cannot meet the core requirements of AR displays. Therefore, the existing BIC and GMR modes cannot simultaneously achieve low-loss transmission in real-world scenes and accurate deflection of virtual images across a wide field of view, limiting the practical application of narrowband nonlocal metasurface gratings in AR optical couplers. Summary of the Invention

[0004] The present invention aims to provide a grating design method and system for optical couplers to solve the above-mentioned technical problems, avoid low grating design efficiency, and realize rapid and automated design of optical coupler gratings.

[0005] To address the aforementioned technical problems, this invention provides a grating design method for an optical coupler, comprising: Obtain the noise vector to be modulated based on the operating wavelength of the optical coupler to be designed; The grating to be optimized is obtained based on the noise vector to be modulated and the pre-trained grating mapping model, and the target grating is obtained based on the grating to be optimized and the pre-trained grating optimization model. The training process of the pre-trained raster mapping model is as follows: Based on the optical coupler, several historical noise vectors are obtained at several preset working wavelengths, and based on the several historical noise vectors and the preset initial mapping model, the initial grating dielectric distribution data and the grating diffraction efficiency at several angles are obtained. Potential dielectric distribution gradient data are obtained based on the grating diffraction efficiency and initial grating dielectric distribution data at several angles. Based on the potential dielectric distribution gradient data, the model parameters of the preset initial mapping model are corrected, and the mapping training index and the training mapping model are obtained. If the mapping training metric does not meet the preset training termination condition, the initial mapping model is updated to the training mapping model, and the training process of the pre-trained raster mapping model is re-executed until the mapping training metric meets the preset training termination condition, and the training mapping model is used as the pre-trained raster mapping model.

[0006] In the above scheme, initial grating dielectric distribution data and grating diffraction efficiencies at several angles are obtained by using several historical noise vectors and a preset initial mapping model. This allows for the rapid acquisition of the initial grating structure and its corresponding optical performance, clarifying the optimization direction of the grating. Next, potential dielectric distribution gradient data is obtained using the initial grating dielectric distribution data and the grating diffraction efficiencies at several angles. Based on this gradient data, the model parameters of the preset initial mapping model are corrected, improving the accuracy and reliability of the grating structure generated by the mapping model. Finally, when the mapping training indicators meet the preset training termination conditions, the trained mapping model is used as a pre-trained grating mapping model, resulting in a mapping model that can be directly used for rapid grating design. This allows the input of the modulation noise vector corresponding to the operating wavelength of the optical coupler to be designed into the pre-trained grating mapping model, quickly generating the grating to be optimized for the optical coupler and avoiding the problem of low grating design efficiency. Finally, the grating to be optimized is input into the pre-trained grating optimization model to obtain the target grating, resulting in a target grating suitable for the optical coupler.

[0007] Furthermore, the step of acquiring several historical noise vectors at several preset operating wavelengths based on the optical coupler, and acquiring initial grating dielectric distribution data and grating diffraction efficiency at several angles based on the several historical noise vectors and a preset initial mapping model, includes: Several wavelength encoding vectors are obtained at several preset operating wavelengths based on an optical coupler; Several wavelength encoded vectors are concatenated with preset structure control components to obtain several historical noise vectors; Several initial latent vectors are obtained based on several historical noise vectors and a preset initial mapping model, and several grating segment width parameters are obtained based on several initial latent vectors; Initial grating dielectric distribution data is obtained based on several grating segment width parameters, and grating diffraction efficiency at several angles is obtained based on the initial grating dielectric distribution data.

[0008] In the above scheme, by acquiring several wavelength-coded vectors, basic data matching the working wavelength can be provided for subsequent noise vector generation. Next, by concatenating these wavelength-coded vectors with preset structural control components, complete input data containing both wavelength and structural control information can be obtained, resulting in several historical noise vectors. Then, using these historical noise vectors and a preset initial mapping model, several potential feature data representing the grating structure, i.e., several initial latent vectors, can be quickly obtained. Subsequently, several grating segment width parameters are obtained from these initial latent vectors, transforming the latent feature data into specific structural parameters usable for grating construction. Next, initial grating dielectric distribution data is obtained from these grating segment width parameters, converting the grating structural parameters into a grating structural distribution usable for electromagnetic simulation. Finally, the grating diffraction efficiency at several angles is obtained from the initial grating dielectric distribution data, yielding the grating's optical performance and providing an evaluation basis for training the initial mapping model.

[0009] Furthermore, the step of obtaining several initial latent vectors based on several historical noise vectors and a preset initial mapping model, and obtaining several grating segment width parameters based on several initial latent vectors, includes: Several initial potential vectors are obtained based on several historical noise vectors and a preset initial mapping model; Several random vectors are obtained based on several initial potential vectors and a pre-defined neural network; Map several random vectors to obtain several mapping ratio coefficient values; Normalize several mapping ratio coefficient values ​​to obtain several intermediate segment variables; Several grating segment width parameters are obtained based on several intermediate segment variables, the preset total grating width, and the preset grating segment width threshold.

[0010] In the above scheme, several initial latent vectors are obtained through several historical noise vectors and a preset initial mapping model, which can quickly obtain several latent feature data representing the grating structure. Next, several random vectors are obtained through these initial latent vectors and a preset neural network, providing basic numerical data for subsequent calculations of grating structure parameters. Then, by mapping these random vectors, several mapping scaling coefficient values ​​are obtained, converting the random vectors into several mapping scaling coefficient values ​​usable for scaling calculations. Subsequently, by normalizing these mapping scaling coefficient values, several intermediate segment variables are obtained, keeping the mapping scaling coefficient values ​​within a reasonable range, facilitating subsequent width calculations. Finally, several grating segment width parameters are obtained through these intermediate segment variables, a preset total grating width, and a preset grating segment width threshold, resulting in stable grating segment width parameters that conform to the preset grating structure constraints.

[0011] Further, the step of obtaining initial grating dielectric distribution data based on several grating segment width parameters, and obtaining grating diffraction efficiency at several angles based on the initial grating dielectric distribution data, includes: The width parameters of several grating segments are mapped and transformed to obtain the initial grating dielectric distribution data; Input the preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver so that the electromagnetic solver can perform simulation based on the preset electromagnetic solver hyperparameters, the preset incident angle array and the initial grating dielectric distribution data, and obtain the grating diffraction efficiency at several angles.

[0012] In the above scheme, by mapping and transforming several grating segment width parameters, the grating structure parameters can be converted into dielectric distribution data that can be used for simulation calculations, thus obtaining the initial grating dielectric distribution data. Next, by inputting the preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver, complete input conditions can be provided for the electromagnetic simulation, ensuring its normal operation. Then, by having the electromagnetic solver perform simulations based on preset electromagnetic solver hyperparameters, preset incident angle arrays, and the initial grating dielectric distribution data, the optical performance of the grating can be accurately calculated, and the grating diffraction efficiency at several angles can be obtained. This provides an evaluation basis for training the grating mapping model, enabling the updated grating mapping model to have multi-angle flat-band joint optimization capabilities when generating grating structures, avoiding the limitation of only being able to generate applicable gratings for fixed angles.

[0013] Furthermore, in obtaining the grating to be optimized based on the noise vector to be modulated and the pre-trained grating mapping model, and in obtaining the target grating based on the grating to be optimized and the pre-trained grating optimization model, the training process of the pre-trained grating optimization model includes: An initial grating is obtained based on a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions. A first grating is then obtained based on the initial grating and a preset initial grating optimization model. The secondary optimization loss value is obtained based on the initial grating and the first grating; The first grating optimization gradient data is obtained based on the secondary optimization loss value; The parameters of the preset initial grating optimization model are optimized using the first grating optimization gradient data to obtain the optimization training index and the grating optimization training model. If the optimized training metrics do not meet the preset iteration termination condition, the initial grating optimization model is updated to the grating optimization training model, and the training process of the pre-trained grating optimization model is re-executed until the optimized training metrics meet the preset iteration termination condition, and the grating optimization training model is used as the pre-trained grating optimization model.

[0014] In the above scheme, an initial grating is obtained through a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions, providing a foundation for subsequent optimization. Next, using the initial grating and the preset initial grating optimization model, the initial grating can be initially optimized to obtain a first grating with improved performance. Then, using the initial grating and the first grating, a secondary optimization loss value is obtained, showing the difference between the optimized and unoptimized grating loss, providing a basis for adjusting the parameters of the grating optimization model. Subsequently, the first grating optimization gradient data is obtained from the secondary optimization loss value. Based on this gradient data, the preset initial grating optimization model is further optimized to improve the grating's diffraction efficiency, resulting in optimized training metrics and a grating optimization training model. If the optimization training metrics do not meet the preset iteration termination condition, the initial grating optimization model is updated to a grating optimization training model and the training process is repeated to continuously optimize the performance of the grating optimization training model, so that the grating structure is continuously optimized in the direction of less loss and higher performance, thereby further improving the diffraction efficiency of the grating and making the grating better meet the requirements of optical couplers. This continues until the optimization training metrics meet the preset iteration termination condition, at which point the grating optimization training model is used as a pre-trained grating optimization model, so that the grating generated by the aforementioned pre-trained grating mapping model can be automatically optimized, realizing the rapid and automated design of optical coupler gratings.

[0015] Further, the step of obtaining an initial grating based on a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions, and obtaining a first grating based on the initial grating and a preset initial grating optimization model, includes: The initial grating is obtained based on a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions; The first potential vector is obtained based on the initial grating; The first grating is obtained based on the first latent vector and the preset initial grating optimization model.

[0016] In the above scheme, by using a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions, an initial grating that meets the requirements can be selected, providing a foundation for subsequent grating optimization. Next, by obtaining the first latent vector from the initial grating, the initial grating can be converted into latent vector data that can be used for model optimization. Then, by obtaining the first grating using a preset initial grating optimization model and the first latent vector, the initial grating can be optimized and adjusted to obtain a first grating with better performance.

[0017] This invention provides a grating design system for an optical coupler, including a module for obtaining the noise vector to be modulated and a target grating generation module, specifically: The module for obtaining the noise vector to be modulated is used to obtain the noise vector to be modulated based on the operating wavelength of the optical coupler to be designed. The target grating generation module is used to obtain the grating to be optimized based on the noise vector to be modulated and the pre-trained grating mapping model, and to obtain the target grating based on the grating to be optimized and the pre-trained grating optimization model. The training process of the pre-trained raster mapping model is as follows: Based on the optical coupler, several historical noise vectors are obtained at several preset working wavelengths, and based on the several historical noise vectors and the preset initial mapping model, the initial grating dielectric distribution data and the grating diffraction efficiency at several angles are obtained. Potential dielectric distribution gradient data are obtained based on the grating diffraction efficiency and initial grating dielectric distribution data at several angles. Based on the potential dielectric distribution gradient data, the model parameters of the preset initial mapping model are corrected, and the mapping training index and the training mapping model are obtained. If the mapping training metric does not meet the preset training termination condition, the initial mapping model is updated to the training mapping model, and the training process of the pre-trained raster mapping model is re-executed until the mapping training metric meets the preset training termination condition, and the training mapping model is used as the pre-trained raster mapping model.

[0018] This invention provides a grating design system for optical couplers. In practical applications, it only requires several historical noise vectors and a preset initial mapping model to obtain initial grating dielectric distribution data and grating diffraction efficiency at several angles. This allows for the rapid acquisition of the initial grating structure and corresponding optical performance, clarifying the optimization direction of the grating. Next, potential dielectric distribution gradient data is obtained using the initial grating dielectric distribution data and grating diffraction efficiency at several angles. Based on this gradient data, the model parameters of the preset initial mapping model are corrected, improving the accuracy and reliability of the grating structure generated by the mapping model. Then, when the mapping training indicators meet the preset training termination conditions, the trained mapping model is used as a pre-trained grating mapping model, resulting in a mapping model that can be directly used for rapid grating design. Therefore, the modulated noise vector corresponding to the operating wavelength of the optical coupler to be designed can be input into the pre-trained grating mapping model using the modulated noise vector acquisition module, quickly generating the grating to be optimized for the optical coupler to be designed, avoiding the problem of low grating design efficiency. Finally, the target grating generation module is used to input the grating to be optimized into the pre-trained grating optimization model to obtain the target grating, thus obtaining a target grating suitable for optical couplers.

[0019] Furthermore, in the target grating generation module, the step of acquiring several historical noise vectors at several preset operating wavelengths based on an optical coupler, and acquiring initial grating dielectric distribution data and grating diffraction efficiency at several angles based on the several historical noise vectors and a preset initial mapping model, includes: Several wavelength encoding vectors are obtained at several preset operating wavelengths based on an optical coupler; Several wavelength encoded vectors are concatenated with preset structure control components to obtain several historical noise vectors; Several initial latent vectors are obtained based on several historical noise vectors and a preset initial mapping model, and several grating segment width parameters are obtained based on several initial latent vectors; Initial grating dielectric distribution data is obtained based on several grating segment width parameters, and grating diffraction efficiency at several angles is obtained based on the initial grating dielectric distribution data.

[0020] In the above scheme, by acquiring several wavelength-coded vectors, basic data matching the working wavelength can be provided for subsequent noise vector generation. Next, by concatenating these wavelength-coded vectors with preset structural control components, complete input data containing both wavelength and structural control information can be obtained, resulting in several historical noise vectors. Then, using these historical noise vectors and a preset initial mapping model, several potential feature data representing the grating structure, i.e., several initial latent vectors, can be quickly obtained. Subsequently, several grating segment width parameters are obtained from these initial latent vectors, transforming the latent feature data into specific structural parameters usable for grating construction. Next, initial grating dielectric distribution data is obtained from these grating segment width parameters, converting the grating structural parameters into a grating structural distribution usable for electromagnetic simulation. Finally, the grating diffraction efficiency at several angles is obtained from the initial grating dielectric distribution data, yielding the grating's optical performance and providing an evaluation basis for training the initial mapping model.

[0021] Furthermore, the step of obtaining several initial latent vectors based on several historical noise vectors and a preset initial mapping model, and obtaining several grating segment width parameters based on several initial latent vectors, includes: Several initial potential vectors are obtained based on several historical noise vectors and a preset initial mapping model; Several random vectors are obtained based on several initial potential vectors and a pre-defined neural network; Map several random vectors to obtain several mapping ratio coefficient values; Normalize several mapping ratio coefficient values ​​to obtain several intermediate segment variables; Several grating segment width parameters are obtained based on several intermediate segment variables, the preset total grating width, and the preset grating segment width threshold.

[0022] In the above scheme, several initial latent vectors are obtained through several historical noise vectors and a preset initial mapping model, which can quickly obtain several latent feature data representing the grating structure. Next, several random vectors are obtained through these initial latent vectors and a preset neural network, providing basic numerical data for subsequent calculations of grating structure parameters. Then, by mapping these random vectors, several mapping scaling coefficient values ​​are obtained, converting the random vectors into several mapping scaling coefficient values ​​usable for scaling calculations. Subsequently, by normalizing these mapping scaling coefficient values, several intermediate segment variables are obtained, keeping the mapping scaling coefficient values ​​within a reasonable range, facilitating subsequent width calculations. Finally, several grating segment width parameters are obtained through these intermediate segment variables, a preset total grating width, and a preset grating segment width threshold, resulting in stable grating segment width parameters that conform to the preset grating structure constraints.

[0023] Further, the step of obtaining initial grating dielectric distribution data based on several grating segment width parameters, and obtaining grating diffraction efficiency at several angles based on the initial grating dielectric distribution data, includes: The width parameters of several grating segments are mapped and transformed to obtain the initial grating dielectric distribution data; Input the preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver so that the electromagnetic solver can perform simulation based on the preset electromagnetic solver hyperparameters, the preset incident angle array and the initial grating dielectric distribution data, and obtain the grating diffraction efficiency at several angles.

[0024] In the above scheme, by mapping and transforming several grating segment width parameters, the grating structure parameters can be converted into dielectric distribution data that can be used for simulation calculations, thus obtaining the initial grating dielectric distribution data. Next, by inputting the preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver, complete input conditions can be provided for the electromagnetic simulation, ensuring its normal operation. Then, by having the electromagnetic solver perform simulations based on preset electromagnetic solver hyperparameters, preset incident angle arrays, and the initial grating dielectric distribution data, the optical performance of the grating can be accurately calculated, and the grating diffraction efficiency at several angles can be obtained. This provides an evaluation basis for training the grating mapping model, enabling the updated grating mapping model to have multi-angle flat-band joint optimization capabilities when generating grating structures, avoiding the limitation of only being able to generate applicable gratings for fixed angles. Attached Figure Description

[0025] To more clearly illustrate the technical solution of this application, 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 from these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating a grating design method for an optical coupler according to an embodiment of the present invention; Figure 2 This is a grating design system architecture diagram of an optical coupler provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, 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.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0029] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0030] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0031] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0032] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0033] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0034] See Figure 1To avoid inefficient grating design and achieve rapid automated design of optical coupler gratings, this embodiment provides a grating design method for optical couplers, including: Step S1: Obtain the noise vector to be modulated based on the operating wavelength of the optical coupler to be designed; Step S2: Obtain the grating to be optimized based on the noise vector to be modulated and the pre-trained grating mapping model, and obtain the target grating based on the grating to be optimized and the pre-trained grating optimization model; The training process of the pre-trained raster mapping model is as follows: Based on the optical coupler, several historical noise vectors are obtained at several preset working wavelengths, and based on the several historical noise vectors and the preset initial mapping model, the initial grating dielectric distribution data and the grating diffraction efficiency at several angles are obtained. Potential dielectric distribution gradient data are obtained based on the grating diffraction efficiency and initial grating dielectric distribution data at several angles. Based on the potential dielectric distribution gradient data, the model parameters of the preset initial mapping model are corrected, and the mapping training index and the training mapping model are obtained. If the mapping training metric does not meet the preset training termination condition, the initial mapping model is updated to the training mapping model, and the training process of the pre-trained raster mapping model is re-executed until the mapping training metric meets the preset training termination condition, and the training mapping model is used as the pre-trained raster mapping model.

[0035] In this embodiment, initial grating dielectric distribution data and grating diffraction efficiencies at several angles are obtained through several historical noise vectors and a preset initial mapping model. This allows for averaging of the grating diffraction efficiencies obtained at multiple angles, quickly yielding an initial grating structure and corresponding optical performance suitable for multi-angle flat-band optimization, thus clarifying the optimization direction of the grating. Next, the grating structure parameters are transformed into pixelated dielectric distribution data of continuous grayscale values ​​suitable for simulation calculations using the initial grating dielectric distribution data and the grating diffraction efficiencies at several angles. This initial grating dielectric distribution data allows for adjusting the grating to gradually evolve from a grayscale distribution to a binary distribution, ensuring the continuity and stability of the subsequent gradient data and avoiding gradient instability during grating generation. Then, the initial dielectric distribution gradient data is transmitted to the grating segment width parameter space through a backpropagation mechanism, ultimately mapping back to the initial latent vector space to obtain latent dielectric distribution gradient data. Based on this gradient data, the preset initial mapping model's parameters are corrected, improving the accuracy and reliability of the grating structure generated by the mapping model.

[0036] Next, when the mapping training metrics meet the preset training termination condition, the trained mapping model is used as a pre-trained grating mapping model, resulting in a mapping model that can be directly used for rapid grating design. This allows the modulation noise vector corresponding to the operating wavelength of the optical coupler to be designed to be input into the pre-trained grating mapping model, quickly generating the grating to be optimized for the optical coupler and avoiding the problem of low grating design efficiency. The mapping training metrics are the number of training iterations and mapping efficiency. The preset training termination condition is that training is terminated when the number of training iterations reaches 300 rounds or the improvement in mapping efficiency does not exceed 5%. Finally, the grating to be optimized is input into the pre-trained grating optimization model to obtain the target grating, resulting in a target grating suitable for the optical coupler.

[0037] Furthermore, the step of acquiring several historical noise vectors at several preset operating wavelengths based on the optical coupler, and acquiring initial grating dielectric distribution data and grating diffraction efficiency at several angles based on the several historical noise vectors and a preset initial mapping model, includes: Several wavelength encoding vectors are obtained at several preset operating wavelengths based on an optical coupler; Several wavelength encoded vectors are concatenated with preset structure control components to obtain several historical noise vectors; Several initial latent vectors are obtained based on several historical noise vectors and a preset initial mapping model, and several grating segment width parameters are obtained based on several initial latent vectors; Initial grating dielectric distribution data is obtained based on several grating segment width parameters, and grating diffraction efficiency at several angles is obtained based on the initial grating dielectric distribution data.

[0038] In this embodiment, by acquiring several wavelength-coded vectors, basic data matching the working wavelength can be provided for subsequent noise vector generation. Next, by concatenating these wavelength-coded vectors with preset structural control components, complete input data containing both wavelength and structural control information can be obtained, resulting in several historical noise vectors. Specifically, the preset structural control components are the minimum grating spacing and the grating height, used to define the minimum spacing and overall thickness of the grating, respectively. Then, using these historical noise vectors and a preset initial mapping model, several potential feature data representing the grating structure, i.e., several initial potential vectors, can be quickly obtained. Subsequently, several grating segment width parameters are obtained using these initial potential vectors, transforming the potential feature data into specific structural parameters usable for grating construction. Next, initial grating dielectric distribution data is obtained using these grating segment width parameters, converting the grating structural parameters into a grating structural distribution usable for electromagnetic simulation. Finally, the grating diffraction efficiency at several angles is obtained using the initial grating dielectric distribution data, providing an evaluation basis for training the initial mapping model.

[0039] Furthermore, the step of obtaining several initial latent vectors based on several historical noise vectors and a preset initial mapping model, and obtaining several grating segment width parameters based on several initial latent vectors, includes: Several initial potential vectors are obtained based on several historical noise vectors and a preset initial mapping model; Several random vectors are obtained based on several initial potential vectors and a pre-defined neural network; Map several random vectors to obtain several mapping ratio coefficient values; Normalize several mapping ratio coefficient values ​​to obtain several intermediate segment variables; Several grating segment width parameters are obtained based on several intermediate segment variables, the preset total grating width, and the preset grating segment width threshold.

[0040] In this embodiment, several initial latent vectors are obtained through several historical noise vectors and a preset initial mapping model, enabling the rapid acquisition of several latent feature data representing the grating structure. The preset initial mapping model is constructed using a pre-normalized (Pre-Norm) multi-layer residual MLP generator network. Before training, the widest hidden layer width of this network is expanded to 256, and the hidden layer depth is increased to 8 layers, with layer widths of 16, 32, 64, 128, 256, 128, 64, 32, and 16 respectively. To avoid gradient vanishing, the network is connected using a multi-layer residual mode, and a preset residual scaling coefficient is introduced. The original residual layer can be expressed as: Initial latent vector = Historical noise vector + (historical noise vector), where (Historical noise vector) is the adjustment amount obtained by inputting the historical noise vector into a preset initial mapping model. Then, several random vectors are obtained through several initial latent vectors and a preset neural network, such as... Where M is the total number of grating segments, providing basic numerical data for subsequent calculations of grating structure parameters. Then, by mapping several random vectors, several mapping scaling coefficient values ​​are obtained, converting the random vectors into several mapping scaling coefficient values ​​usable for scaling calculations. Specifically, this involves performing a Sigmoid mapping on several random vectors. ,in, For the i-th mapping scaling factor value, For random vectors, this mapping process can transform unrestricted... The data is compressed to the interval (0, 1). Subsequently, by normalizing several mapping scaling coefficient values, several intermediate segment variables are obtained. This ensures that the mapping scaling coefficient values ​​remain within a reasonable range, facilitating subsequent width calculations. Specifically: ,in, Let i be the intermediate segment variable. This is the upper limit of the mapping ratio coefficient. =1, This is the lower limit of the mapping scaling factor. =0. Finally, by using several intermediate segment variables, a preset total grating width, and a preset grating segment width threshold, several grating segment width parameters are obtained, resulting in stable grating segment width parameters that conform to the preset grating structure constraints. Specifically: ,in, For the grating segment width parameter, The preset grating segment width threshold is defined by L, which is the preset total grating width. This ensures that the width of each grating segment naturally meets the threshold. And the total width of the grating is always L.

[0041] Further, the step of obtaining initial grating dielectric distribution data based on several grating segment width parameters, and obtaining grating diffraction efficiency at several angles based on the initial grating dielectric distribution data, includes: The width parameters of several grating segments are mapped and transformed to obtain the initial grating dielectric distribution data; Input the preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver so that the electromagnetic solver can perform simulation based on the preset electromagnetic solver hyperparameters, the preset incident angle array and the initial grating dielectric distribution data, and obtain the grating diffraction efficiency at several angles.

[0042] In this embodiment, initial grating dielectric distribution data can be obtained by mapping and transforming several grating segment width parameters. Then, by inputting a preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver, complete input conditions can be provided for the electromagnetic simulation, ensuring its normal operation. Since the electromagnetic solver only supports a single angle input at a time, this embodiment uses the input angle as a loop variable to repeatedly call the preset single-angle calculation sub-function. By enabling the electromagnetic solver to perform simulation based on preset electromagnetic solver hyperparameters, the preset incident angle array, and the initial grating dielectric distribution data, the optical performance of the grating can be accurately calculated, and the grating diffraction efficiency at several angles can be obtained. This provides an evaluation basis for training the grating mapping model, enabling the updated grating mapping model to have multi-angle flat-band joint optimization capabilities when generating grating structures, avoiding the limitation of only generating applicable gratings for fixed angles. The preset electromagnetic solver hyperparameter is -1, which can adjust the conventional 0th-order diffraction secondary of the optical coupler to a -1 diffraction secondary.

[0043] Furthermore, in obtaining the grating to be optimized based on the noise vector to be modulated and the pre-trained grating mapping model, and in obtaining the target grating based on the grating to be optimized and the pre-trained grating optimization model, the training process of the pre-trained grating optimization model includes: An initial grating is obtained based on a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions. A first grating is then obtained based on the initial grating and a preset initial grating optimization model. The secondary optimization loss value is obtained based on the initial grating and the first grating; The first grating optimization gradient data is obtained based on the secondary optimization loss value; The parameters of the preset initial grating optimization model are optimized using the first grating optimization gradient data to obtain the optimization training index and the grating optimization training model. If the optimized training metrics do not meet the preset iteration termination condition, the initial grating optimization model is updated to the grating optimization training model, and the training process of the pre-trained grating optimization model is re-executed until the optimized training metrics meet the preset iteration termination condition, and the grating optimization training model is used as the pre-trained grating optimization model.

[0044] In this embodiment, an initial grating is obtained through a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions, providing a foundation for subsequent optimization. The preset vector selection conditions are: selecting the grating with the highest diffraction efficiency from several gratings generated from several historical noise vectors as the initial grating. Next, the initial grating is preliminarily optimized using the initial grating and the preset initial grating optimization model to obtain a first grating with improved performance. The preset initial grating optimization model uses the Adam optimizer. Then, by inputting the initial grating and the first grating into the electromagnetic solver, the diffraction efficiency of the initial grating and the first grating are obtained. The secondary optimization loss value is then obtained using the initial grating diffraction efficiency and the first grating diffraction efficiency, providing a basis for adjusting the parameters of the grating optimization model. Finally, the gradient in the current computational space is cleared to prevent the gradient of the current iteration from overlapping with the gradient of the previous iteration, while improving computational efficiency.

[0045] Subsequently, the gradient of the first grating diffraction efficiency with respect to the potential vector is calculated in reverse using the second-order optimization loss value, i.e., the first grating optimization gradient data. Based on the first grating optimization gradient data, the parameters of the preset initial grating optimization model are optimized, which can further improve the diffraction efficiency of the grating at the target diffraction order (-1 order), and obtain the optimized training index and the grating optimization training model. In order to avoid numerical instability in the second-order optimization stage of the grating, such as gradient explosion leading to excessively large update step size, efficiency oscillation or even divergence, the L2 norm is introduced to prune the gradient: when the L2 norm of the current gradient vector exceeds 20, the gradient is scaled proportionally so that its corresponding L2 norm is limited to within 20, thereby significantly improving the convergence stability. If the optimized training metrics do not meet the preset iteration termination condition, the initial grating optimization model is updated to a grating optimization training model, and the training process is repeated to continuously optimize the performance of the grating optimization training model. This allows the grating structure to be continuously optimized towards lower loss and higher performance, thereby further improving the diffraction efficiency of the grating and making it better meet the requirements of optical couplers. This continues until the optimized training metrics meet the preset iteration termination condition. At this point, the grating optimization training model is used as a pre-trained grating optimization model, enabling automatic optimization of the gratings generated by the aforementioned pre-trained grating mapping model, achieving rapid and automated design of optical coupler gratings. The preset iteration termination condition is that training terminates when the number of iterations reaches 600 rounds.

[0046] After each training of the grating optimization model, a periodic evaluation mechanism is set up: the current iteration number in the grating optimization process is obtained and the evaluation frequency in the preset simulation configuration parameters is obtained. The current iteration number is determined by the preset remainder operation to determine whether the current iteration number meets the evaluation conditions, that is, the remainder between the current iteration number and the evaluation frequency is calculated. If the remainder is equal to 0, the subsequent evaluation process is triggered; if the remainder is not equal to 0, the current evaluation is skipped and the next round of iteration is continued.

[0047] After entering the evaluation process, physical simulation is performed based on the updated first grating and electromagnetic solver without affecting the optimization calculation. The reflection efficiency of the first grating under the -1st order diffraction secondary is calculated, and the calculated reflection efficiency is uniformly converted into a standard numerical format to eliminate format differences for easy recording and comparison.

[0048] Finally, the converted standard efficiency values ​​are stored in a pre-set efficiency storage list. As iterations continue, multiple sets of efficiency data will be gradually recorded. Based on all the data recorded in this list, the performance changes during the secondary optimization process can be viewed intuitively, thereby analyzing the optimization effect and convergence trend.

[0049] The grating designed in this embodiment employs a specific structural arrangement. Through the mutual coupling between nonlocal guided modes, the shift in resonant wavelength with the incident angle is offset, forming a flat-band resonant characteristic within a certain angular range. This allows the grating to maintain a essentially constant resonant wavelength over a wide incident angle range (50°~68°), while simultaneously maintaining high and uniform diffraction efficiency. It can stably tune the light at the target wavelength (565nm) to the -1 diffraction order, thus better meeting the requirements of AR optical couplers for wide field of view, narrow band response, and flat-band characteristics. It can be applied to free-space light AR devices, significantly reducing the optical path volume of the AR optical coupler while effectively reducing optical path complexity.

[0050] Further, the step of obtaining an initial grating based on a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions, and obtaining a first grating based on the initial grating and a preset initial grating optimization model, includes: The initial grating is obtained based on a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions; The first potential vector is obtained based on the initial grating; The first grating is obtained based on the first latent vector and the preset initial grating optimization model.

[0051] In this embodiment, by using a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions, an initial grating that meets the requirements can be selected, providing a foundation for subsequent grating optimization. Next, a first latent vector is obtained from the initial grating, converting it into latent vector data suitable for model optimization. Then, a first grating is obtained using a preset initial grating optimization model and the first latent vector, allowing for optimization and adjustment of the initial grating to obtain a first grating with better performance.

[0052] This embodiment provides a grating design system for an optical coupler. Please refer to [link to relevant documentation]. Figure 2 It includes a module for obtaining the noise vector to be modulated and a module for generating the target grating, specifically: The module for obtaining the noise vector to be modulated is used to obtain the noise vector to be modulated based on the operating wavelength of the optical coupler to be designed. The target grating generation module is used to obtain the grating to be optimized based on the noise vector to be modulated and the pre-trained grating mapping model, and to obtain the target grating based on the grating to be optimized and the pre-trained grating optimization model. The training process of the pre-trained raster mapping model is as follows: Based on the optical coupler, several historical noise vectors are obtained at several preset working wavelengths, and based on the several historical noise vectors and the preset initial mapping model, the initial grating dielectric distribution data and the grating diffraction efficiency at several angles are obtained. Potential dielectric distribution gradient data are obtained based on the grating diffraction efficiency and initial grating dielectric distribution data at several angles. Based on the potential dielectric distribution gradient data, the model parameters of the preset initial mapping model are corrected, and the mapping training index and the training mapping model are obtained. If the mapping training metric does not meet the preset training termination condition, the initial mapping model is updated to the training mapping model, and the training process of the pre-trained raster mapping model is re-executed until the mapping training metric meets the preset training termination condition, and the training mapping model is used as the pre-trained raster mapping model.

[0053] This embodiment provides a grating design system for optical couplers. In practical applications, it only requires obtaining initial grating dielectric distribution data and grating diffraction efficiency at several angles using several historical noise vectors and a preset initial mapping model. This allows for the rapid acquisition of the initial grating structure and corresponding optical performance, clarifying the optimization direction of the grating. Next, potential dielectric distribution gradient data is obtained using the initial grating dielectric distribution data and grating diffraction efficiency at several angles. Based on this gradient data, the model parameters of the preset initial mapping model are corrected, improving the accuracy and reliability of the grating structure generated by the mapping model. Then, when the mapping training index meets the preset training termination condition, the trained mapping model is used as a pre-trained grating mapping model, resulting in a mapping model that can be directly used for rapid grating design. Therefore, the modulated noise vector corresponding to the operating wavelength of the optical coupler to be designed can be input into the pre-trained grating mapping model using the modulated noise vector acquisition module, quickly generating the grating to be optimized for the optical coupler to be designed, avoiding the problem of low grating design efficiency. Finally, the target grating generation module is used to input the grating to be optimized into the pre-trained grating optimization model to obtain the target grating, thus obtaining a target grating suitable for optical couplers.

[0054] Furthermore, in the target grating generation module, the step of acquiring several historical noise vectors at several preset operating wavelengths based on an optical coupler, and acquiring initial grating dielectric distribution data and grating diffraction efficiency at several angles based on the several historical noise vectors and a preset initial mapping model, includes: Several wavelength encoding vectors are obtained at several preset operating wavelengths based on an optical coupler; Several wavelength encoded vectors are concatenated with preset structure control components to obtain several historical noise vectors; Several initial latent vectors are obtained based on several historical noise vectors and a preset initial mapping model, and several grating segment width parameters are obtained based on several initial latent vectors; Initial grating dielectric distribution data is obtained based on several grating segment width parameters, and grating diffraction efficiency at several angles is obtained based on the initial grating dielectric distribution data.

[0055] In this embodiment, by acquiring several wavelength-coded vectors, basic data matching the operating wavelength can be provided for subsequent noise vector generation. Next, by concatenating these wavelength-coded vectors with preset structural control components, complete input data containing both wavelength and structural control information can be obtained, resulting in several historical noise vectors. Then, using these historical noise vectors and a preset initial mapping model, several potential feature data representing the grating structure, i.e., several initial potential vectors, can be quickly obtained. Subsequently, several grating segment width parameters are obtained using these initial potential vectors, transforming the potential feature data into specific structural parameters usable for grating construction. Next, initial grating dielectric distribution data is obtained using these grating segment width parameters, converting the grating structural parameters into a grating structural distribution usable for electromagnetic simulation. Finally, the grating diffraction efficiency at several angles is obtained using the initial grating dielectric distribution data, yielding the grating's optical performance and providing an evaluation basis for training the initial mapping model.

[0056] Furthermore, the step of obtaining several initial latent vectors based on several historical noise vectors and a preset initial mapping model, and obtaining several grating segment width parameters based on several initial latent vectors, includes: Several initial potential vectors are obtained based on several historical noise vectors and a preset initial mapping model; Several random vectors are obtained based on several initial potential vectors and a pre-defined neural network; Map several random vectors to obtain several mapping ratio coefficient values; Normalize several mapping ratio coefficient values ​​to obtain several intermediate segment variables; Several grating segment width parameters are obtained based on several intermediate segment variables, the preset total grating width, and the preset grating segment width threshold.

[0057] In this embodiment, several initial latent vectors are obtained using several historical noise vectors and a preset initial mapping model, enabling the rapid acquisition of several latent feature data representing the grating structure. Next, several random vectors are obtained using these initial latent vectors and a preset neural network, providing basic numerical data for subsequent calculations of grating structure parameters. Then, by mapping these random vectors, several mapping scaling coefficient values ​​are obtained, converting the random vectors into several mapping scaling coefficient values ​​usable for scaling calculations. Subsequently, by normalizing these mapping scaling coefficient values, several intermediate segment variables are obtained, keeping the mapping scaling coefficient values ​​within a reasonable range, facilitating subsequent width calculations. Finally, several grating segment width parameters are obtained using these intermediate segment variables, a preset total grating width, and a preset grating segment width threshold, resulting in stable grating segment width parameters that conform to preset grating structure constraints.

[0058] Further, the step of obtaining initial grating dielectric distribution data based on several grating segment width parameters, and obtaining grating diffraction efficiency at several angles based on the initial grating dielectric distribution data, includes: The width parameters of several grating segments are mapped and transformed to obtain the initial grating dielectric distribution data; Input the preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver so that the electromagnetic solver can perform simulation based on the preset electromagnetic solver hyperparameters, the preset incident angle array and the initial grating dielectric distribution data, and obtain the grating diffraction efficiency at several angles.

[0059] In this embodiment, by mapping and transforming several grating segment width parameters, the grating structure parameters can be converted into dielectric distribution data that can be used for simulation calculations, thus obtaining initial grating dielectric distribution data. Next, by inputting a preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver, complete input conditions can be provided for the electromagnetic simulation, ensuring its normal operation. Then, by having the electromagnetic solver perform simulations based on preset electromagnetic solver hyperparameters, preset incident angle arrays, and the initial grating dielectric distribution data, the optical performance of the grating can be accurately calculated, and the grating diffraction efficiency at several angles can be obtained. This provides an evaluation basis for training the grating mapping model, enabling the updated grating mapping model to have multi-angle flat-band joint optimization capabilities when generating grating structures, avoiding the limitation of only being able to generate applicable gratings for fixed angles.

[0060] This embodiment also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the functions of the system as described above.

[0061] It is understood that the above system embodiments correspond to the method embodiments of the present invention, and can implement the grating design method for an optical coupler provided by any of the above method embodiments of the present invention.

[0062] It should be noted that the system embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0063] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A grating design method for an optical coupler, characterized in that, include: Obtain the noise vector to be modulated based on the operating wavelength of the optical coupler to be designed; The grating to be optimized is obtained based on the noise vector to be modulated and the pre-trained grating mapping model, and the target grating is obtained based on the grating to be optimized and the pre-trained grating optimization model. The training process of the pre-trained raster mapping model is as follows: Based on the optical coupler, several historical noise vectors are obtained at several preset working wavelengths, and based on the several historical noise vectors and the preset initial mapping model, the initial grating dielectric distribution data and the grating diffraction efficiency at several angles are obtained. Potential dielectric distribution gradient data are obtained based on the grating diffraction efficiency and initial grating dielectric distribution data at several angles. Based on the potential dielectric distribution gradient data, the model parameters of the preset initial mapping model are corrected, and the mapping training index and the training mapping model are obtained. If the mapping training metric does not meet the preset training termination condition, the initial mapping model is updated to the training mapping model, and the training process of the pre-trained raster mapping model is re-executed until the mapping training metric meets the preset training termination condition, and the training mapping model is used as the pre-trained raster mapping model.

2. The grating design method for an optical coupler according to claim 1, characterized in that, The process of acquiring several historical noise vectors at several preset operating wavelengths based on an optical coupler, and acquiring initial grating dielectric distribution data and grating diffraction efficiency at several angles based on the several historical noise vectors and a preset initial mapping model, includes: Several wavelength encoding vectors are obtained at several preset operating wavelengths based on an optical coupler; Several wavelength encoded vectors are concatenated with preset structure control components to obtain several historical noise vectors; Several initial latent vectors are obtained based on several historical noise vectors and a preset initial mapping model, and several grating segment width parameters are obtained based on several initial latent vectors; Initial grating dielectric distribution data is obtained based on several grating segment width parameters, and grating diffraction efficiency at several angles is obtained based on the initial grating dielectric distribution data.

3. The grating design method for an optical coupler according to claim 2, characterized in that, The process of obtaining several initial latent vectors based on several historical noise vectors and a preset initial mapping model, and obtaining several grating segment width parameters based on several initial latent vectors, includes: Several initial potential vectors are obtained based on several historical noise vectors and a preset initial mapping model; Several random vectors are obtained based on several initial potential vectors and a pre-defined neural network; Map several random vectors to obtain several mapping ratio coefficient values; Normalize several mapping ratio coefficient values ​​to obtain several intermediate segment variables; Several grating segment width parameters are obtained based on several intermediate segment variables, the preset total grating width, and the preset grating segment width threshold.

4. The grating design method for an optical coupler according to claim 2, characterized in that, The process of obtaining initial grating dielectric distribution data based on several grating segment width parameters, and obtaining grating diffraction efficiency at several angles based on the initial grating dielectric distribution data, includes: The width parameters of several grating segments are mapped and transformed to obtain the initial grating dielectric distribution data; Input the preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver so that the electromagnetic solver can perform simulation based on the preset electromagnetic solver hyperparameters, the preset incident angle array and the initial grating dielectric distribution data, and obtain the grating diffraction efficiency at several angles.

5. The grating design method for an optical coupler according to claim 1, characterized in that, In obtaining the grating to be optimized based on the noise vector to be modulated and a pre-trained grating mapping model, and in obtaining the target grating based on the grating to be optimized and the pre-trained grating optimization model, the training process of the pre-trained grating optimization model includes: An initial grating is obtained based on a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions. A first grating is then obtained based on the initial grating and a preset initial grating optimization model. The secondary optimization loss value is obtained based on the initial grating and the first grating; The first grating optimization gradient data is obtained based on the secondary optimization loss value; The parameters of the preset initial grating optimization model are optimized using the first grating optimization gradient data to obtain the optimization training index and the grating optimization training model. If the optimized training metrics do not meet the preset iteration termination condition, the initial grating optimization model is updated to the grating optimization training model, and the training process of the pre-trained grating optimization model is re-executed until the optimized training metrics meet the preset iteration termination condition, and the grating optimization training model is used as the pre-trained grating optimization model.

6. The grating design method for an optical coupler according to claim 5, characterized in that, The process of obtaining an initial grating based on a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions, and obtaining a first grating based on the initial grating and a preset initial grating optimization model, includes: The initial grating is obtained based on a pre-trained grating mapping model, several historical noise vectors, and preset vector selection conditions; The first potential vector is obtained based on the initial grating; The first grating is obtained based on the first latent vector and the preset initial grating optimization model.

7. A grating design system for an optical coupler, characterized in that, It includes a module for obtaining the noise vector to be modulated and a module for generating the target grating, specifically: The module for obtaining the noise vector to be modulated is used to obtain the noise vector to be modulated based on the operating wavelength of the optical coupler to be designed. The target grating generation module is used to obtain the grating to be optimized based on the noise vector to be modulated and the pre-trained grating mapping model, and to obtain the target grating based on the grating to be optimized and the pre-trained grating optimization model. The training process of the pre-trained raster mapping model is as follows: Based on the optical coupler, several historical noise vectors are obtained at several preset working wavelengths, and based on the several historical noise vectors and the preset initial mapping model, the initial grating dielectric distribution data and the grating diffraction efficiency at several angles are obtained. Potential dielectric distribution gradient data are obtained based on the grating diffraction efficiency and initial grating dielectric distribution data at several angles. Based on the potential dielectric distribution gradient data, the model parameters of the preset initial mapping model are corrected, and the mapping training index and the training mapping model are obtained. If the mapping training metric does not meet the preset training termination condition, the initial mapping model is updated to the training mapping model, and the training process of the pre-trained raster mapping model is re-executed until the mapping training metric meets the preset training termination condition, and the training mapping model is used as the pre-trained raster mapping model.

8. The grating design system for an optical coupler according to claim 7, characterized in that, In the target grating generation module, the step of acquiring several historical noise vectors at several preset operating wavelengths based on an optical coupler, and acquiring initial grating dielectric distribution data and grating diffraction efficiency at several angles based on the several historical noise vectors and a preset initial mapping model, includes: Several wavelength encoding vectors are obtained at several preset operating wavelengths based on an optical coupler; Several wavelength encoded vectors are concatenated with preset structure control components to obtain several historical noise vectors; Several initial latent vectors are obtained based on several historical noise vectors and a preset initial mapping model, and several grating segment width parameters are obtained based on several initial latent vectors; Initial grating dielectric distribution data is obtained based on several grating segment width parameters, and grating diffraction efficiency at several angles is obtained based on the initial grating dielectric distribution data.

9. A grating design system for an optical coupler according to claim 8, characterized in that, The process of obtaining several initial latent vectors based on several historical noise vectors and a preset initial mapping model, and obtaining several grating segment width parameters based on several initial latent vectors, includes: Several initial potential vectors are obtained based on several historical noise vectors and a preset initial mapping model; Several random vectors are obtained based on several initial potential vectors and a pre-defined neural network; Map several random vectors to obtain several mapping ratio coefficient values; Normalize several mapping ratio coefficient values ​​to obtain several intermediate segment variables; Several grating segment width parameters are obtained based on several intermediate segment variables, the preset total grating width, and the preset grating segment width threshold.

10. A grating design system for an optical coupler according to claim 8, characterized in that, The process of obtaining initial grating dielectric distribution data based on several grating segment width parameters, and obtaining grating diffraction efficiency at several angles based on the initial grating dielectric distribution data, includes: The width parameters of several grating segments are mapped and transformed to obtain the initial grating dielectric distribution data; Input the preset incident angle array and the initial grating dielectric distribution data into the electromagnetic solver so that the electromagnetic solver can perform simulation based on the preset electromagnetic solver hyperparameters, the preset incident angle array and the initial grating dielectric distribution data, and obtain the grating diffraction efficiency at several angles.

Images