Optical device modeling method and device based on sparse grid space mapping

Abstract

The invention provides an optical device modeling method and device based on sparse grid space mapping, and the method comprises the steps: carrying out the simulation of geometric design parameters, input end optical parameters and output end optical parameters of a target optical device based on different precisions, and constructing fine grid data and sparse grid data; and an artificial neural network model is trained to construct a rough network, and the geometric design parameters and the input end optical parameters are mapped to the output end optical parameters, so that the optical response can be quickly fitted, and the calculation amount is reduced. And furthermore, the output of the rough model is mapped to fine grid data by establishing a space mapping module, so that the output of high-precision optical response is realized, and the modeling efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of optical device modeling, and in particular to an optical device modeling method and device based on sparse grid space mapping. Background Art

[0002] In the process of optical device design and optimization, traditional methods usually require nested scanning in multiple geometric dimensions and at different wavelengths, which consumes a lot of time because repeated fine-mesh simulations are required. Faced with complex optical structures, traditional physical modeling methods are not only computationally expensive, but also have low modeling efficiency, making it difficult to meet the needs of rapid design and optimization. At the same time, the need for a large amount of high-precision simulation data further increases the consumption of time and resources, limiting the optimization and design progress of optical devices.

[0003] In addition, traditional modeling methods are often difficult to effectively handle when faced with the coupling of multiple physical fields, resulting in insufficient model prediction accuracy. As the complexity of optical device design increases, how to efficiently build accurate models and reduce the required training data and computing costs has become a scientific problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides an optical device modeling method and apparatus based on sparse grid space mapping to eliminate or improve one or more defects existing in the prior art, namely, the problem that the optical device modeling calculation is complex, consumes a large amount of computing resources and cannot meet the accuracy requirements.

[0005] One aspect of the present invention provides an optical device modeling method based on sparse grid space mapping, the method comprising the following steps: Simulating geometric design parameters, input end optical parameters and output end optical parameters of a target optical device based on different accuracies to construct fine grid data and sparse grid data, wherein the simulation accuracy of the fine grid data is higher than that of the sparse grid data; Using the sparse grid data to train a preset artificial neural network, and establish a rough model for mapping the geometric design parameters and the input end optical parameters to the output end optical parameters; constructing a spatial mapping module to map the output of the coarse model to the fine grid data; The coarse model is combined with the spatial mapping module as a surrogate model of the target optical device, wherein the coarse model provides an optical response for rapid prediction and the spatial mapping module is used to provide an optical response for accurate prediction.

[0006] In some embodiments, the method further includes: sampling the fine grid data to obtain sparse grid data, wherein the sampling form includes random sampling, grid downsampling, Latin hypercube sampling, orthogonal sampling, Sobol sequence sampling, kernel density estimation sampling or regional importance sampling.

[0007] In some embodiments, using the sparse grid data to train a preset artificial neural network comprises the following steps: Preprocessing the sparse grid data includes normalizing and standardizing the data; dividing the sparse grid data into a training set, a validation set, and a test set according to a set ratio; and performing data expansion and enhancement by interpolation and noise disturbance; The artificial neural network is trained using the preprocessed sparse grid data, wherein the artificial neural network comprises an input layer, a hidden layer and an output layer, the number of nodes in the input layer is equal to the number of input parameters, the hidden layer is provided with at least 3 layers, and the number of neurons in each layer is 2 to 4 times the input parameters; the output layer adopts a fully connected layer; the hidden layer and the output layer are provided with a linear activation function layer; the parameters of the artificial neural network model are updated by constructing a mean square error loss to obtain the rough model.

[0008] In some embodiments, the artificial neural network is trained using a fully connected neural network, a convolutional neural network, or a recurrent neural network.

[0009] In some embodiments, the method further comprises: The parameters of the artificial neural network are updated using Adam or RMSProp optimizer, and L2 regularization or Dropout is added to prevent overfitting.

[0010] In some embodiments, the method further comprises: Fine grid data is constructed based on optical devices with multiple other geometric design parameters that are the same type as the target optical device, and the sparse grid data is sampled.

[0011] In some embodiments, the spatial mapping module uses multi-source linear regression or support vector machine regression to establish a mapping relationship between the output of the coarse model and the fine grid data.

[0012] On the other hand, the present invention also provides an optical device modeling device based on sparse grid space mapping, comprising a processor, a memory, and a computer program / instructions stored in the memory, wherein the processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the device implements the steps of the above method.

[0013] On the other hand, the present invention further provides a computer-readable storage medium having a computer program / instruction stored thereon, which implements the steps of the above method when the computer program / instruction is executed by a processor.

[0014] On the other hand, the present invention also provides a computer program product, comprising a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0015] The beneficial effects of the present invention are at least: The optical device modeling method and device based on sparse grid space mapping of the present invention simulates the geometric design parameters, input end optical parameters and output end optical parameters of the target optical device based on different precisions to construct fine grid data and sparse grid data, trains the artificial neural network model to construct a rough network to map the geometric design parameters and the input end optical parameters to the output end optical parameters, can quickly fit the optical response and reduce the amount of calculation. Further, by establishing a spatial mapping module to map the output of the rough model to the fine grid data, the output of the high-precision optical response is achieved, and the efficiency of modeling is improved.

[0016] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention may be achieved and obtained by the structures specifically indicated in the specification and the accompanying drawings.

[0017] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings: Figure 1 The figure is a flow chart of an optical device modeling method based on sparse grid space mapping according to an embodiment of the present invention.

[0019] Figure 2 It is a logical schematic diagram of an optical device modeling method based on sparse grid space mapping according to another embodiment of the present invention.

[0020] Figure 3 This is a logical schematic diagram of an optical device modeling method based on space mapping technology and machine learning according to another embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0022] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0023] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0024] When it comes to large-scale, multi-domain optical device modeling and optimization tasks, traditional optical device modeling methods have problems such as insufficient modeling accuracy, low optimization efficiency, and poor adaptability. Traditional methods rely on high-precision meshing, which results in huge computational complexity and is difficult to apply to complex optical system modeling and optimization.

[0025] The present invention provides an optical device modeling method based on sparse grid space mapping, such as Figure 1 and Figure 2 As shown, the method includes the following steps S101-S104: Step S101: Based on different accuracies, the geometric design parameters, input end optical parameters and output end optical parameters of the target optical device are simulated to construct fine grid data and sparse grid data, wherein the simulation accuracy of the fine grid data is higher than that of the sparse grid data.

[0026] Step S102: using the sparse grid data to train a preset artificial neural network, and establishing a rough model for mapping geometric design parameters and input end optical parameters to output end optical parameters.

[0027] Step S103: construct a spatial mapping module to map the output of the coarse model to fine grid data.

[0028] Step S104: combining the rough model and the spatial mapping module into a substitute model of the target optical device, wherein the rough model provides an optical response for rapid prediction, and the spatial mapping module is used to provide an optical response for accurate prediction.

[0029] The present invention is a modeling for optical devices. Exemplarily, optical devices include but are not limited to optical waveguides, optical modulators and sensors. The geometric design parameters of these optical devices determine their characteristics during the processing of optical signals, and the input optical parameters refer to the physical characteristic parameters of the input optical signal when the optical device performs the corresponding processing task; the output optical parameters refer to the physical characteristic parameters of the output optical signal after the optical device acts. Exemplarily, for optical waveguides, geometric parameters may include the width and height of the core, the thickness of the cladding, the length of the waveguide, the bending radius, the spacing of the waveguide array, the refractive index of the material, the dispersion characteristics, the absorption coefficient, etc. The input optical parameters may include: input light wavelength, input light mode, input light power and incident angle. The output optical parameters may include output mode distribution, optical power, mode conversion efficiency, transmission loss, polarization characteristics, etc.

[0030] In step S101, fine grid data can be obtained by actual measurement or data simulation. For experimental measurement, the performance data of optical devices are collected by high-precision test equipment (such as a spectrum analyzer, an interferometer, or a near-field scanning optical microscope). First, prepare sample optical devices that meet the design requirements and ensure that their structure and material properties are precisely controllable. During the experiment, adjust the parameters of the input light (such as wavelength, power, polarization state, etc.), gradually scan all optical characteristic parameters of interest, and record the response of the output light (such as transmittance, reflectivity, mode distribution, etc.). To ensure the spatial resolution of the data, a micro-nano manipulation device can be selected, and the measurement accuracy and repeatability can be optimized. Another way is numerical simulation, using high-precision numerical simulation tools (such as finite element method FEM, finite difference time domain method FDTD, etc.) to generate data. Before simulation, a detailed physical model needs to be prepared, including the geometric structure, material properties, and boundary conditions of the device. Then, set the fine grid division, and the grid unit size is determined according to the target accuracy (such as 5nm). During the simulation process, the input parameters are scanned step by step to obtain the output results under different conditions. To improve data accuracy, a grid convergence test is required to verify that the simulation error is below the preset threshold. In addition, to ensure the credibility of the simulation results, a small amount of experimental data can be compared with the simulation results for calibration and verification. Through these two methods, high-precision, reliable fine grid data can be obtained, providing basic support for subsequent model construction and optimization.

[0031] In some embodiments, in the sparse grid data obtained by sampling the fine grid data, the sampling form includes random sampling, grid downsampling, Latin hypercube sampling, orthogonal sampling, Sobol sequence sampling, kernel density estimation sampling or regional importance sampling.

[0032] Random sampling refers to randomly selecting sample points from fine grid data. It is simple to implement, but it is only applicable when the data is uniformly distributed.

[0033] Grid downsampling refers to downsampling the original fine grid according to a certain interval or ratio. It is simple and intuitive and suitable for regular grid data.

[0034] Latin Hypercube Sampling (LHS) divides each dimension into several intervals, randomly selects a sample point from each interval to form the final sample set, and can uniformly sample in multidimensional space, control the number of samples and improve diversity.

[0035] Orthogonal sampling uses orthogonal experimental design to select representative sample points from the design space. Its wide coverage can ensure that the sample points are evenly distributed in the parameter space.

[0036] Sobol sequence sampling is based on quasi-random sequence generation points, with the goal of uniform coverage of space. It has good coverage effect on high-dimensional space, strong controllability, and is suitable for scenarios that require uniform distribution.

[0037] Kernel density estimation sampling uses kernel density estimation to obtain data distribution and performs sampling according to density distribution. It can better capture the local characteristics of data distribution.

[0038] Regional importance sampling prioritizes sampling of areas of interest or areas where the objective function changes dramatically based on prior knowledge or objective functions. This can enhance coverage of key areas and improve sampling efficiency.

[0039] In some embodiments, the method further comprises: constructing fine grid data based on optical devices with multiple other geometric design parameters of the same type as the target optical device, and sampling sparse grid data. In order to ensure data diversity and improve the generalization ability of the rough model finally obtained, when constructing fine grid data and sparse grid data, optical devices with multiple other geometric design parameters of the same type as the target optical device can be used to jointly experiment or simulate to generate data.

[0040] In step S102, the preset artificial neural network is trained using sparse grid data, including the following steps S201-S202: Step S201: preprocessing the sparse grid data, the preprocessing includes normalizing and standardizing the data; dividing the sparse grid data into a training set, a validation set and a test set according to a set ratio; and expanding and enhancing the data through interpolation and noise disturbance.

[0041] Step S202: The preprocessed sparse grid data is used to train the artificial neural network, wherein the artificial neural network comprises an input layer, a hidden layer and an output layer, wherein the number of nodes in the input layer is equal to the number of input parameters, the hidden layer is provided with at least 3 layers, and the number of neurons in each layer is 2 to 4 times of the input parameters; the output layer adopts a fully connected layer; the hidden layer and the output layer are provided with a linear activation function layer; the parameters of the artificial neural network model are updated by constructing a mean square error loss to obtain a rough model.

[0042] Furthermore, in step S201, the input parameters and output data are normalized or standardized (e.g., the data is scaled to [0, 1] or the mean is 0 and the variance is 1). The data set is divided into a training set (70% to 80%), a validation set (10% to 15%) and a test set (10% to 15%).

[0043] In some embodiments, the artificial neural network is trained using a fully connected neural network, a convolutional neural network, or a recurrent neural network.

[0044] In some embodiments, the method further comprises: updating parameters of the artificial neural network using an Adam or RMSProp optimizer, and adding L2 regularization or Dropout to prevent overfitting.

[0045] In other embodiments, if the input and output are linearly related, a linear regression model can be used. For input and output with nonlinear relationships, a polynomial regression model can be used to expand the expressive power of the linear regression model by adding polynomial terms. Support vector machine regression can also be used to capture nonlinear relationships through kernel functions, which can handle high-dimensional and nonlinear problems and has good generalization capabilities for small data sets.

[0046] In some embodiments, the spatial mapping module uses multi-source linear regression or support vector machine regression to establish a mapping relationship between the output of the coarse model and the fine grid data.

[0047] In some embodiments, both the coarse model and the spatial mapping model are trained using NeuroModelerPlus software to achieve high-precision modeling and optimization in multiple physical fields. The training accuracy of the fine grid data is 5nm to ensure that the model can provide high-precision prediction results.

[0048] Specifically, the spatial mapping relationship between the output of the coarse model and the fine grid data is achieved by establishing a mapping model (such as a mapping neural network or regression model), the purpose of which is to adjust the coarse prediction to a high-precision result that is closer to the fine grid data. The following is a specific implementation plan: 1. Data preparation Collect coarse model outputs, using the coarse model trained on the sparse grid data to generate predicted outputs as input data. These outputs are usually of lower accuracy but cover the main features of the design space.

[0049] Collect fine mesh data and use numerical simulation (such as FEM, FDTD) or experimental measurement to obtain high-resolution and accurate output data. Fine mesh data provides high-precision target values ​​and is a reference for mapping relationship optimization.

[0050] Data matching,pairs the output of the coarse model with the corresponding fine grid data one by one to form input-output sample pairs for training the mapping model.

[0051] 2. Selection of mapping model The multivariate linear regression model is used when the relationship between the coarse model output and the fine grid data is close to linear. The model form can be: Among them, w is the weight matrix, b is the bias vector, Alternatively, a support vector machine (SVM) is used for mapping relationships with nonlinear characteristics, and a kernel function (such as a Gaussian kernel or a polynomial kernel) is selected to map the input into a high-dimensional space to fit the nonlinear relationship.

[0052] Alternatively, a mapping neural network is used, which is suitable for scenarios where the mapping relationship is highly nonlinear or complex. The input layer is used to input the output of the rough model, and the hidden layer has 1 to 3 layers, using nonlinear activation functions (such as ReLU). The output layer outputs fine grid prediction values. This form is highly flexible and can capture complex nonlinear mappings. The mean square error (MSE) loss function is used, expressed as: ; Where N represents the sample data size, represents fine grid data, is the model's prediction for the fine-grid data.

[0053] On the other hand, the present invention also provides an optical device modeling device based on sparse grid space mapping, comprising a processor, a memory, and a computer program / instructions stored in the memory, wherein the processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the device implements the steps of the above method.

[0054] The embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the aforementioned edge computing server deployment method are implemented. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0055] The following is a description of a specific embodiment: This embodiment provides an optical device modeling method based on space mapping (SM) technology and machine learning. Figure 3 As shown, it includes: traditional optical device design method and machine learning-based optimization design method.

[0056] In one embodiment of the present embodiment, an optical device replacement model combining artificial neural network and space mapping technology includes a coarse model and a fine model. The coarse model is trained by artificial neural network, and the coarse model is corrected by space mapping technology, so as to obtain a high-precision optical device replacement model. The structure of the space mapping parameter model is proposed, including a coarse model, a fine model and a space mapping parameter model. By training the mapping parameters of the coarse model and the fine model, a mapping relationship between the coarse model output and the fine grid data is established, so as to achieve the optimization of the optical device design.

[0057] The training mechanism includes model training data collection, model training and model correction. By collecting a large amount of training data and continuously training and correcting the model, a high-precision overall parameter model is eventually obtained.

[0058] The development process includes initial model establishment, model training, model correction and model verification. By replacing and correcting the parameters between the coarse model and the fine model, a high-precision optical device design model is finally obtained.

[0059] The optical device modeling and optimization method based on sparse grid space mapping technology described in this embodiment realizes efficient modeling and optimization of optical devices. The specific implementation method is as follows: Example 1: Sparse grid data sampling Sparse grid data sampling is performed in the design space of the optical device to obtain initial sample data. The diversity and representativeness of the samples are ensured by the orthogonal distribution method. The specific operation steps are as follows: 1.1 Determine the design space: According to the structure and functional requirements of the optical equipment, determine the boundaries and ranges of the physical parameters contained in the design space.

[0060] 1.2 Sparse grid division: Perform sparse grid division in the design space to obtain a number of sample points.

[0061] 1.3 Sample data collection: Perform physical simulation or experimental testing at each sample point to obtain the corresponding sample data.

[0062] Example 2: Building a rough model Artificial neural networks are used to learn sparse grid data and build a rough model of the optical device. The specific steps are as follows: 2.1 Data preprocessing: Standardize the sparse grid data to eliminate the dimensional differences of the data.

[0063] 2.2 Network structure design: Design a suitable artificial neural network structure, including the number of nodes and connection methods of the input layer, hidden layer and output layer.

[0064] 2.3 Network training: Use the back propagation algorithm to train the neural network and optimize the network's weights and bias parameters.

[0065] 2.4 Model verification: Use some sample data to verify the rough model and evaluate the prediction accuracy and generalization ability of the model.

[0066] Example 3: Establishing spatial mapping relationship Based on the spatial mapping relationship between the coarse model and the fine grid data, the optical device model is further optimized. The specific steps are as follows: 3.1 Data preparation: Collect high-precision grid data as the basic data for establishing spatial mapping relationships.

[0067] 3.2 Mapping relationship modeling: Use multiple linear regression, support vector machine or other appropriate methods to establish the mapping relationship between the coarse model output and the high-precision grid data.

[0068] 3.3 Model optimization: Use the mapping relationship to optimize the rough model and improve the prediction accuracy of the model.

[0069] Example 4: High-precision substitution model construction Based on the established spatial mapping relationship, a high-precision replacement model of the optical device is constructed using high-precision grid data. The specific steps are as follows: 4.1 Network structure optimization: According to the characteristics of high-precision grid data, the neural network structure is optimized.

[0070] 4.2 Model training: Use an improved training algorithm to train the neural network and build a high-precision alternative model.

[0071] 4.3 Model verification: Some high-precision grid data are used to verify the alternative model and evaluate the prediction accuracy and stability of the model.

[0072] Therefore, the optical device modeling and optimization method based on sparse grid space mapping technology described in this embodiment achieves efficient modeling and optimization of optical devices through sparse grid data sampling, rough model construction, spatial mapping relationship establishment, high-precision alternative model construction and multi-physical field response prediction, improves the accuracy of the model and optimization efficiency, and is suitable for modeling and optimization tasks of complex optical systems.

[0073] The embodiment of the present invention further provides a computer device, which may include a processor, a memory and an image acquisition device, wherein the processor and the memory may be connected via a bus or other means, and this embodiment takes the bus connection as an example. The image acquisition device may be connected to the processor and the memory via a wired or wireless manner.

[0074] The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0075] The memory is a non-transitory computer-readable storage medium that can be used to store non-transitory software programs, non-transitory computer executable programs and modules. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory, that is, the image color correction method in the above method embodiment is implemented.

[0076] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required by at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the storage may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0077] The one or more modules are stored in the memory and when executed by the processor, perform the following steps: Figure 1-Figure 3 Method in the embodiment shown.

[0078] As an implementation method, the functions of the receiver and the transmitter in the present invention may be implemented by a transceiver circuit or a dedicated chip for transceiver, and the processor may be implemented by a dedicated processing chip, a processing circuit or a general-purpose chip.

[0079] As another implementation, it is possible to use a general-purpose computer to implement the authentication device and authentication server provided in the embodiment of the present invention, that is, the program code for implementing the functions of the processor, receiver, and transmitter is stored in a memory, and the general-purpose processor implements the functions of the processor, receiver, and transmitter by executing the code in the memory.

[0080] In summary, the optical device modeling method and device based on sparse grid space mapping of the present invention simulates the geometric design parameters, input end optical parameters and output end optical parameters of the target optical device based on different precisions to construct fine grid data and sparse grid data, trains the artificial neural network model to construct a coarse network to map the geometric design parameters and the input end optical parameters to the output end optical parameters, can quickly fit the optical response and reduce the amount of calculation. Further, by establishing a spatial mapping module to map the output of the coarse model to the fine grid data, the output of the high-precision optical response is achieved, and the efficiency of modeling is improved.

[0081] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0082] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0083] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An optical device modeling method based on sparse grid space mapping, characterized in that: The method comprises the following steps: Simulating geometric design parameters, input end optical parameters and output end optical parameters of a target optical device based on different accuracies to construct fine grid data and sparse grid data, wherein the simulation accuracy of the fine grid data is higher than that of the sparse grid data; Using the sparse grid data to train a preset artificial neural network, and establish a rough model for mapping the geometric design parameters and the input end optical parameters to the output end optical parameters; constructing a spatial mapping module to map the output of the coarse model to the fine grid data; The coarse model is combined with the spatial mapping module as a surrogate model of the target optical device, wherein the coarse model provides an optical response for rapid prediction and the spatial mapping module is used to provide an optical response for accurate prediction.

2. The optical device modeling method based on sparse grid space mapping according to claim 1, characterized in that: The method further includes: sampling the fine grid data to obtain sparse grid data, wherein the sampling forms include random sampling, grid downsampling, Latin hypercube sampling, orthogonal sampling, Sobol sequence sampling, kernel density estimation sampling or regional importance sampling.

3. The optical device modeling method based on sparse grid space mapping according to claim 1, characterized in that: Using the sparse grid data to train a preset artificial neural network includes the following steps: Preprocessing the sparse grid data, wherein the preprocessing includes normalizing and standardizing the data; Dividing the sparse grid data into a training set, a validation set and a test set according to a set ratio; and performing data expansion and enhancement by interpolation and noise disturbance; The artificial neural network is trained using the preprocessed sparse grid data, wherein the artificial neural network comprises an input layer, a hidden layer and an output layer, the number of nodes in the input layer is equal to the number of input parameters, the hidden layer is provided with at least 3 layers, and the number of neurons in each layer is 2 to 4 times the input parameters; the output layer adopts a fully connected layer; the hidden layer and the output layer are provided with a linear activation function layer; the parameters of the artificial neural network model are updated by constructing a mean square error loss to obtain the rough model.

4. The optical device modeling method based on sparse grid space mapping according to claim 3 is characterized in that: The artificial neural network is trained using a fully connected neural network, a convolutional neural network or a recurrent neural network.

5. The optical device modeling method based on sparse grid space mapping according to claim 4, characterized in that: The method further comprises: The parameters of the artificial neural network are updated using Adam or RMSProp optimizer, and L2 regularization or Dropout is added to prevent overfitting.

6. The optical device modeling method based on sparse grid space mapping according to claim 1, characterized in that: The method further comprises: Fine grid data is constructed based on optical devices with multiple other geometric design parameters that are the same type as the target optical device, and the sparse grid data is sampled.

7. The optical device modeling method based on sparse grid space mapping according to claim 1, characterized in that: The spatial mapping module uses multi-source linear regression or support vector machine regression to establish a mapping relationship between the output of the coarse model and the fine grid data.

8. An optical device modeling device based on sparse grid space mapping, comprising a processor, a memory, and a computer program / instruction stored in the memory, characterized in that: The processor is used to execute the computer program / instructions. When the computer program / instructions are executed, the device implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method as claimed in any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images