Metamaterial absorber design method and system based on natural language processing and loop prediction network

Abstract

The invention discloses a metamaterial absorber design method and system based on a natural language processing and loop prediction network, and belongs to the field of nanometer photonics device design. The method comprises the following steps: acquiring initial sample data containing metamaterial absorber geometric parameters and corresponding optical response, and training a loop prediction network model to establish a function relationship between the two; performing preprocessing, feature mapping and length calibration on the unstructured design demand text through a natural language processing module to generate a structured feature vector; and inputting the feature vectors into a trained model, and outputting geometric parameters through a loop prediction mode (based on residual feedback optimization) so as to complete on-demand design of the metamaterial absorber. The corresponding system comprises a data acquisition and model training module, a natural language processing module, a loop prediction network module and an on-demand design module. According to the method, the design efficiency can be remarkably improved, the design precision and generalization ability are improved, the design threshold is lowered, the computing resource consumption is reduced, and engineering popularization of the metamaterial absorber is facilitated.

Description

Technical Field

[0001] This invention relates to the field of nanophotonic device design, and in particular to a metamaterial absorber design method and system based on natural language processing and recurrent prediction networks. Background Technology

[0002] Metamaterial absorbers, due to their tunable electromagnetic properties, have shown great application potential in optoelectronic devices, energy harvesting, and biosensing. Among them, metamaterial absorbers in the visible to near-infrared bands have become a research hotspot because they cover the main frequency band of solar radiation and the optical window of biological tissues. Their core advantage lies in achieving efficient light absorption through subwavelength unit structures, which can significantly improve the energy conversion efficiency of solar cells and enhance the detection sensitivity of sensors. However, the performance of metamaterial absorbers is strongly coupled with their structural parameters. The geometry, material properties, and periodic arrangement of the unit structure all affect their absorption characteristics, leading to a highly complex design process. Traditional design methods rely on trial and error or finite element simulation iterations, which not only consume a large amount of computational resources but also struggle to optimize absorption characteristics across wide bands, multiple frequency bands, or specific angles. Especially in multi-parameter collaborative optimization scenarios, the design cycle often lasts for weeks or even months, severely restricting the engineering application of metamaterial absorbers. With the development of artificial intelligence technology, machine learning algorithms have provided new ideas for materials design. Natural language processing (NLP) technology can parse key parameters in design requirement texts, enabling the transformation from unstructured descriptions to structured design variables. Recurrent prediction networks (RNNs) can model the nonlinear mapping relationship between structural parameters and absorption performance, capturing the impact of parameter changes on broadband absorption characteristics through time-series learning. While some existing research has applied neural networks to metamaterial performance prediction, a complete design system integrating requirement analysis, parameter optimization, and performance prediction has not yet been established. Furthermore, in broadband absorption optimization across the visible to near-infrared bands, issues such as insufficient accuracy and weak generalization ability remain, making it difficult to meet the demands of practical applications for rapid and intelligent design. Summary of the Invention

[0003] The purpose of this invention is to overcome one or more shortcomings of the prior art and provide a metamaterial absorber design method and system based on natural language processing and recurrent prediction networks.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A method for designing metamaterial absorbers based on natural language processing and recurrent prediction networks includes the following steps:

[0006] Step 1: Obtain initial sample data of the metamaterial absorber, which includes the geometric parameters of the metamaterial absorber and its corresponding optical response data; use the initial sample data to train a recurrent prediction network model to establish a functional relationship between the geometric parameters of the metamaterial absorber and its optical response;

[0007] Step 2: Process the unstructured design requirement text input by the user through the natural language processing module to generate structured feature vectors;

[0008] Step 3: Input the feature vector into the trained recurrent prediction network model, and output the geometric parameters of the metamaterial absorber through the recurrent prediction mode;

[0009] Step 4: Based on the geometric parameters output by the cyclic prediction network model, complete the on-demand design of the metamaterial absorber.

[0010] Furthermore, in step 1, different metamaterial absorber structures are constructed and their optical responses are characterized using the time-domain finite element difference method. The optical response is the absorptivity, covering a wavelength range from visible light to near-infrared light. The initial sample data is generated by calculation using finite element simulation software.

[0011] Furthermore, in step 2, the natural language processing module processes the design requirement text, including text data preprocessing operations, which are cleaning and denoising, word segmentation and stop word removal, and standardization.

[0012] The text cleaning and denoising process removes special characters, repeated sentences, and erroneous encoding data. The word segmentation and stop word removal process uses Markov models or conditional random fields for word segmentation and removes words from a pre-defined stop word list containing prepositions and auxiliary words. The standardization process converts the text into a word frequency-inverse document frequency vector or a word embedding vector.

[0013] Furthermore, in step 2, the natural language processing module's processing of the design requirement text also includes feature mapping operations and length calibration operations; the feature mapping operation includes constructing a domain dictionary based on a professional literature corpus in the field of metamaterials, expanding the vector through a word vector generation algorithm using a context window, and concatenating the processed vectors; the length calibration operation calibrates the length of the feature vector to a preset dimension by supplementing data with zero padding or a normal distribution.

[0014] Furthermore, in step 1, when training the recurrent prediction network model, multiple decision trees are used to construct an ensemble model; before training, the geometric parameters of all metamaterial absorbers are linearly normalized to values ​​between [0, 1].

[0015] Furthermore, in step 3, the iteration number of the cyclic prediction mode is a preset number. In each iteration, the predicted absorption spectrum is calculated based on the current geometric parameters, the predicted absorption spectrum is compared with the ideal absorption spectrum to obtain the residual, and the residual is fed back to the next prediction input until the residual is less than a preset threshold or the preset number of iterations is reached. The objective function of the cyclic prediction mode is the weighted sum of the output results of multiple decision trees, and the weight of the weighted sum is determined by minimizing the mean square error between the predicted absorption spectrum and the ideal absorption spectrum.

[0016] Furthermore, in step 4, the specific process of on-demand design is as follows: the design requirements of the target metamaterial absorber are input into the system in text form. The system transmits the text to the natural language processing module, which processes it and converts it into a feature vector that can be read by the recurrent prediction network model. The model predicts the corresponding geometric parameters based on the feature vector, and then constructs the target metamaterial absorber structure.

[0017] A metamaterial absorber design system based on natural language processing and recurrent prediction networks includes:

[0018] The data acquisition and model training module is used to construct the metamaterial absorber structure and characterize its optical response through the temporal finite element difference method, acquire initial sample data, and train the recurrent prediction network model using the initial sample data.

[0019] The natural language processing module receives unstructured design requirement text input by the user, performs text cleaning and noise reduction, word segmentation and stop word removal, standardization, feature mapping and length calibration to generate structured feature vectors;

[0020] The recurrent prediction network module is used to receive the feature vector output by the natural language processing module and output the geometric parameters of the metamaterial absorber through the recurrent prediction mode;

[0021] The on-demand design module is used to construct and output the target metamaterial absorber structure based on the geometric parameters output by the cyclic prediction network module.

[0022] Furthermore, the natural language processing module includes a text data preprocessing submodule, a feature mapping submodule, and a length calibration submodule;

[0023] The text data preprocessing submodule performs cleaning and noise reduction, word segmentation and stop word removal, and standardization operations.

[0024] The feature mapping submodule performs dictionary construction, vector expansion, and vector concatenation operations;

[0025] The length calibration submodule performs a feature vector length calibration operation, calibrating the feature vector length to 201 dimensions.

[0026] Furthermore, the recurrent prediction network module includes a model training submodule and a recurrent prediction submodule;

[0027] The model training submodule trains a multi-decision-tree ensemble model based on a dataset containing metamaterial structural parameters and corresponding absorption spectra.

[0028] The cyclic prediction submodule takes the feature vector as input, adopts a many-to-one prediction method, and outputs the corresponding geometric parameters iteratively based on the ideal absorption spectrum through a cyclic prediction mode with an objective function and a preset number of iterations.

[0029] The beneficial effects of this invention are:

[0030] (1) Significantly improve design efficiency: The design of a wideband absorber, which traditionally requires several weeks to complete, can be completed within hours by this system through iterative optimization of a cyclic predictive network. The design cycle is shortened by more than 90%, and no manual intervention is required, achieving full automation of the process.

[0031] (2) Improve design accuracy and generalization ability: The GBDT integrated model can accurately capture the nonlinear relationship between structural parameters and absorption performance, and reduce the error in prediction in the visible to near-infrared band; at the same time, through the domain dictionary and context expansion mechanism, the system can handle diverse design requirement texts.

[0032] (3) Lowering the design threshold: The natural language processing module converts unstructured requirements into structured parameters. Users do not need to have professional metamaterial design knowledge. They can complete the design simply by describing it in natural language, which significantly lowers the technical application threshold and is conducive to the engineering promotion of metamaterial absorbers.

[0033] (4) Reduced computational resource consumption: Compared with the traditional finite element simulation iterative method, this system replaces part of the simulation calculation with machine learning model, which reduces the computational resource consumption of a single design and saves hardware costs. Attached Figure Description

[0034] Figure 1 An exemplary schematic diagram of a metamaterial absorber is provided for this embodiment;

[0035] Figure 2 This is a top view of the metamaterial absorber provided in this embodiment;

[0036] Figure 3 This is a flowchart illustrating the natural language processing and recurrent prediction network system architecture during the specific implementation of this embodiment.

[0037] Figure 4 This is a flowchart illustrating the implementation of this embodiment;

[0038] Figure 5This is a graph showing the data collected by the time-domain difference finite element method in this embodiment;

[0039] Figure 6 This is a loss diagram for iterative prediction during the specific implementation of this embodiment;

[0040] Figure 7 This is a diagram illustrating the effect of parameter design during the specific implementation of this embodiment;

[0041] Figure 8 This embodiment illustrates the error prediction process of nanophotonic devices during specific implementation.

[0042] Figure 9 This diagram illustrates the effect of on-demand design and verification of the recurrent prediction network in this embodiment.

[0043] Figure 10 This diagram illustrates the effect of constructing performance measurement of nanophotonic devices during the specific implementation of this embodiment. Detailed Implementation

[0044] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] Example 1

[0046] A method for designing metamaterial absorbers based on natural language processing and recurrent prediction networks includes the following steps:

[0047] Step 1: Obtain initial sample data of the metamaterial absorber, which includes the geometric parameters of the metamaterial absorber and its corresponding optical response data; use the initial sample data to train a recurrent prediction network model to establish a functional relationship between the geometric parameters of the metamaterial absorber and its optical response;

[0048] Step 2: Process the unstructured design requirement text input by the user through the natural language processing module to generate structured feature vectors;

[0049] Step 3: Input the feature vector into the trained recurrent prediction network model, and output the geometric parameters of the metamaterial absorber through the recurrent prediction mode;

[0050] Step 4: Based on the geometric parameters output by the cyclic prediction network model, complete the on-demand design of the metamaterial absorber.

[0051] In step 1, different metamaterial absorber structures are constructed and their optical responses are characterized using the time-domain finite element difference method. The optical response is the absorptivity, covering a wavelength range from visible light to near-infrared light. The initial sample data is generated by finite element simulation software.

[0052] In step 2, the natural language processing module processes the design requirement text, including text data preprocessing operations, which are cleaning and denoising, word segmentation and stop word removal, and standardization.

[0053] The text cleaning and denoising process removes special characters, repeated sentences, and erroneous encoding data. The word segmentation and stop word removal process uses Markov models or conditional random fields for word segmentation and removes words from a pre-defined stop word list containing prepositions and auxiliary words. The standardization process converts the text into a word frequency-inverse document frequency vector or a word embedding vector.

[0054] In step 2, the natural language processing module's processing of the design requirement text also includes feature mapping and length calibration operations. The feature mapping operation includes constructing a domain dictionary based on a professional literature corpus in the field of metamaterials, expanding the vectors through a word vector generation algorithm using a context window, and concatenating the processed vectors. The length calibration operation calibrates the length of the feature vectors to a preset dimension by supplementing the data with zero padding or a normal distribution.

[0055] In step 1, when training the recurrent prediction network model, multiple decision trees are used to construct an ensemble model; before training, the geometric parameters of all metamaterial absorbers are linearly normalized to values ​​between [0, 1].

[0056] In step 3, the iteration number of the cyclic prediction mode is a preset number. In each iteration, the predicted absorption spectrum is calculated based on the current geometric parameters, the predicted absorption spectrum is compared with the ideal absorption spectrum to obtain the residual, and the residual is fed back to the next prediction input until the residual is less than a preset threshold or the preset number of iterations is reached. The objective function of the cyclic prediction mode is the weighted sum of the output results of multiple decision trees. The weight of the weighted sum is determined by minimizing the mean square error between the predicted absorption spectrum and the ideal absorption spectrum.

[0057] In step 4, the specific process of on-demand design is as follows: the design requirements of the target metamaterial absorber are input into the system in text form. The system transmits the text to the natural language processing module, which processes it and converts it into a feature vector that can be read by the recurrent prediction network model. The model predicts the corresponding geometric parameters based on the feature vector, and then constructs the target metamaterial absorber structure.

[0058] A metamaterial absorber design system based on natural language processing and recurrent prediction networks includes:

[0059] The data acquisition and model training module is used to construct the metamaterial absorber structure and characterize its optical response through the temporal finite element difference method, acquire initial sample data, and train the recurrent prediction network model using the initial sample data.

[0060] The natural language processing module receives unstructured design requirement text input by the user, performs text cleaning and noise reduction, word segmentation and stop word removal, standardization, feature mapping and length calibration to generate structured feature vectors;

[0061] The recurrent prediction network module is used to receive the feature vector output by the natural language processing module and output the geometric parameters of the metamaterial absorber through the recurrent prediction mode;

[0062] The on-demand design module is used to construct and output the target metamaterial absorber structure based on the geometric parameters output by the cyclic prediction network module.

[0063] The natural language processing module includes a text data preprocessing submodule, a feature mapping submodule, and a length calibration submodule.

[0064] The text data preprocessing submodule performs cleaning and noise reduction, word segmentation and stop word removal, and standardization operations.

[0065] The feature mapping submodule performs dictionary construction, vector expansion, and vector concatenation operations;

[0066] The length calibration submodule performs a feature vector length calibration operation, calibrating the feature vector length to 201 dimensions.

[0067] The recurrent prediction network module includes a model training submodule and a recurrent prediction submodule.

[0068] The model training submodule trains a multi-decision-tree ensemble model based on a dataset containing metamaterial structural parameters and corresponding absorption spectra.

[0069] The cyclic prediction submodule takes the feature vector as input, adopts a many-to-one prediction method, and outputs the corresponding geometric parameters iteratively based on the ideal absorption spectrum through a cyclic prediction mode with an objective function and a preset number of iterations.

[0070] Example 2

[0071] In this embodiment, the device is designed using a silicon wafer as the substrate, gold as the main material of the absorber (a metal with a high melting point and stability, and good plasmon resonance properties), aluminum oxide as the dielectric layer, and gold as the material of the top structure. Figure 1As shown, the metamaterial absorber provided in this embodiment includes a metal-dielectric-metal layer. The light source is required to be planar, with vertical incidence from the top of the unit structure, and the wavelength range of the light source is from visible to near-infrared light. The distance between the light source and the device is greater than one wavelength. The boundary conditions for the simulation interval are periodic boundary conditions in the x and y directions, and a perfectly matched layer in the z direction. A reflectivity monitor is placed above the light source to obtain the reflectivity distribution within the simulated wavelength range. Other areas are set as air. Since the substrate is a silicon wafer, the transmittance is almost zero, and a transmittance monitor is not required; therefore, the absorptivity value is A = 1 - RT, T = 0. Figure 2 As shown in the top view diagram, it illustrates the structure of different absorbers, covering the visible to near-infrared light bands.

[0072] After the required simulation conditions are set, such as Figure 5 As shown, the initial dataset is formed by assembling the range of geometric parameters or arranging them according to a pattern to ensure reasonable coverage. In this specific implementation, there are 100,000 initial sample data label values. The absorption spectra corresponding to the geometric parameters are obtained through numerical simulation using the finite element difference method. To enable the model to better learn the mapping relationship between the input data X and Y, the label values ​​need to be normalized. This specific implementation uses linear normalization, i.e.:

[0073] ;

[0074] like Figure 3 As shown in Figure 4, the natural language processing module converts the unstructured design requirement text input by the user into a structured feature vector, providing input parameters for the recurrent prediction network. For example, the phrase "Design a metamaterial absorber with an absorptivity of 90% at 660nm" retains its core requirement parameters after cleaning. This needs to be converted into data values ​​processed by the recurrent prediction network to generate a normal distribution map showing an absorptivity of 0.9 at 660nm in the 400-700nm band, i.e.:

[0075] ;

[0076] The recurrent prediction network employs multiple gradient boosting decision trees to construct an ensemble model. Each decision tree has a depth of 8 layers and 32 leaf nodes. One hundred decision trees are trained sequentially using a forward distribution algorithm. Each new tree is trained based on the prediction residuals of all previous trees, gradually reducing the prediction error. A 5-fold cross-validation method is used for model training. The training set is divided into 5 subsets, with 4 subsets selected for training and 1 subset for validation in each iteration. Model performance is optimized by adjusting the learning rate (initial value 0.1, decaying by 10% every 100 rounds) and the regularization coefficient (L2 regularization parameter set to 0.01). The final model's mean squared error on the validation set is controlled within 0.005. During training, to improve prediction accuracy, a recurrent prediction mode is used in the system: a recurrent optimization strategy with 5000 iterations is employed, with each iteration including performance prediction, residual calculation, and parameter adjustment.

[0077] In this specific implementation example, Figure 6 The loss value predicted by the model decreases rapidly within the first 50 steps and stabilizes after 1000 steps, with a stable loss value of 3.35E-5. Figure 7 The geometric parameters output by the inverse network are shown, and it can be seen that the predicted parameters have a small error compared to the actual parameters. Figure 8 As shown, the fitting coefficients for each absorption point in the absorption spectrum are all greater than 0.95. Figure 9 The flowchart shown illustrates the on-demand device design process. For example, to design a 1530nm metamaterial absorber with an absorptivity exceeding 0.9, the 201 absorption points obtained through natural language processing are input into a recurrent prediction network. The model, pre-trained, constructs multiple objective functions, and the output geometric parameters are obtained. The device is then fabricated based on these parameters. Before fabrication, the absorption spectrum can be calculated using the finite element method (FEM). The on-demand absorption spectrum is compared, and the obtained geometric parameters are used for calculation using the finite time-difference method. The calculated results are then compared with the input. The flowchart includes the input absorption spectrum data, the predicted geometric parameters, and the simulated absorption spectrum. Figure 10 The performance metrics of the demonstrated design device show high accuracy, with the fitting coefficients for each absorption point in the absorption spectrum all exceeding 0.95. The results clearly demonstrate that this system is beneficial for the assisted design of metamaterial absorbers and enables on-demand design based on the user's desired optical response. This offers valuable insights, especially given the rapid development of metamaterial absorbers and the ever-increasing demand for high-performance device design. The system in this embodiment provides strong support and reference value for the future development of self-designed metamaterial absorbers.

[0078] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for designing metamaterial absorbers based on natural language processing and recurrent prediction networks, characterized in that, Includes the following steps: Step 1: Obtain initial sample data of the metamaterial absorber, which includes the geometric parameters of the metamaterial absorber and its corresponding optical response data; use the initial sample data to train a recurrent prediction network model to establish a functional relationship between the geometric parameters of the metamaterial absorber and its optical response; Step 2: Process the unstructured design requirement text input by the user through the natural language processing module to generate structured feature vectors; Step 3: Input the feature vector into the trained recurrent prediction network model, and output the geometric parameters of the metamaterial absorber through the recurrent prediction mode; Step 4: Based on the geometric parameters output by the cyclic prediction network model, complete the on-demand design of the metamaterial absorber.

2. The metamaterial absorber design method based on natural language processing and recurrent prediction networks according to claim 1, characterized in that, In step 1, different metamaterial absorber structures are constructed and their optical responses are characterized using the time-domain finite element difference method. The optical response is the absorptivity, covering a wavelength range from visible light to near-infrared light. The initial sample data is generated by finite element simulation software.

3. The metamaterial absorber design method based on natural language processing and recurrent prediction networks according to claim 1, characterized in that, In step 2, the natural language processing module processes the design requirement text, including text data preprocessing operations, which are cleaning and denoising, word segmentation and stop word removal, and standardization. The text cleaning and denoising process removes special characters, repeated sentences, and erroneous encoding data. The word segmentation and stop word removal process uses Markov models or conditional random fields for word segmentation and removes words from a pre-defined stop word list containing prepositions and auxiliary words. The standardization process converts the text into a word frequency-inverse document frequency vector or a word embedding vector.

4. The metamaterial absorber design method based on natural language processing and recurrent prediction networks according to claim 3, characterized in that, In step 2, the natural language processing module's processing of the design requirement text also includes feature mapping operations and length calibration operations; the feature mapping operations include constructing a domain dictionary based on a professional literature corpus in the field of metamaterials, expanding vectors through a word vector generation algorithm using a context window, and concatenating the processed vectors; The length calibration operation calibrates the length of the feature vector to a preset dimension by supplementing data with zero padding or a normal distribution.

5. The metamaterial absorber design method based on natural language processing and recurrent prediction networks according to claim 1, characterized in that, In step 1, when training the recurrent prediction network model, multiple decision trees are used to construct an ensemble model; before training, the geometric parameters of all metamaterial absorbers are linearly normalized to values ​​between [0, 1].

6. The metamaterial absorber design method based on natural language processing and recurrent prediction networks according to claim 1, characterized in that, In step 3, the iteration number of the cyclic prediction mode is a preset number. In each iteration, the predicted absorption spectrum is calculated based on the current geometric parameters, the predicted absorption spectrum is compared with the ideal absorption spectrum to obtain the residual, and the residual is fed back to the next prediction input until the residual is less than a preset threshold or the preset number of iterations is reached. The objective function of the cyclic prediction mode is the weighted sum of the output results of multiple decision trees. The weight of the weighted sum is determined by minimizing the mean square error between the predicted absorption spectrum and the ideal absorption spectrum.

7. The metamaterial absorber design method based on natural language processing and recurrent prediction networks according to claim 1, characterized in that, In step 4, the specific process of on-demand design is as follows: the design requirements of the target metamaterial absorber are input into the system in text form. The system transmits the text to the natural language processing module, which processes it and converts it into a feature vector that can be read by the recurrent prediction network model. The model predicts the corresponding geometric parameters based on the feature vector, and then constructs the target metamaterial absorber structure.

8. A metamaterial absorber design system based on natural language processing and recurrent prediction networks, characterized in that, include: The data acquisition and model training module is used to construct the metamaterial absorber structure and characterize its optical response through the temporal finite element difference method, acquire initial sample data, and train the recurrent prediction network model using the initial sample data. The natural language processing module receives unstructured design requirement text input by the user, performs text cleaning and noise reduction, word segmentation and stop word removal, standardization, feature mapping and length calibration to generate structured feature vectors; The recurrent prediction network module is used to receive the feature vector output by the natural language processing module and output the geometric parameters of the metamaterial absorber through the recurrent prediction mode; The on-demand design module is used to construct and output the target metamaterial absorber structure based on the geometric parameters output by the cyclic prediction network module.

9. The metamaterial absorber design system based on natural language processing and recurrent prediction networks according to claim 8, characterized in that, The natural language processing module includes a text data preprocessing submodule, a feature mapping submodule, and a length calibration submodule; The text data preprocessing submodule performs cleaning and noise reduction, word segmentation and stop word removal, and standardization operations. The feature mapping submodule performs dictionary construction, vector expansion, and vector concatenation operations; The length calibration submodule performs a feature vector length calibration operation, calibrating the feature vector length to 201 dimensions.

10. The metamaterial absorber design system based on natural language processing and recurrent prediction networks according to claim 8, characterized in that, The recurrent prediction network module includes a model training submodule and a recurrent prediction submodule; The model training submodule trains a multi-decision-tree ensemble model based on a dataset containing metamaterial structural parameters and corresponding absorption spectra. The cyclic prediction submodule takes the feature vector as input, adopts a many-to-one prediction method, and outputs the corresponding geometric parameters iteratively based on the ideal absorption spectrum through a cyclic prediction mode with an objective function and a preset number of iterations.

Images