Optical system design method and system based on knowledge base retrieval enhancement large model

Abstract

The invention discloses an optical system design method and system based on a knowledge base retrieval enhancement large model, and belongs to the technical field of optical design and artificial intelligence. The method comprises the following steps: constructing an optical system knowledge base containing vectorized coded optical lens design data; a data pair is formed based on an optical system design requirement and a corresponding parameter list of each surface of the optical system, a training data set is constructed, the training data set is utilized and a low-rank adaptive technology is adopted to train a large language model, and then the trained large language model is combined with an optical system knowledge base through a knowledge base retrieval enhancement module to obtain an optical system knowledge base. Obtaining a knowledge base retrieval enhancement large model; and inputting the design requirement of the to-be-designed optical system into the knowledge base to retrieve the enhanced large model to obtain optical system structure parameters meeting the design requirement of the to-be-designed optical system, namely obtaining a parameter list of each surface of the optical system. The method can effectively improve the optical design efficiency, and is suitable for the design of various optical systems.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical design automation, and in particular relates to an optical system design method and system based on knowledge base retrieval and enhanced large model. Background Art

[0002] Optical systems are core components in numerous high-tech fields, including imaging, lighting, and sensing. The traditional optical system design process relies heavily on the expertise and accumulated experience of design engineers. It typically involves selecting an initial structure based on requirements, performing multiple rounds of parameter optimization, and evaluating image quality using optical design software (such as Zemax and CODE V). This process is not only time-consuming and labor-intensive, but also often makes it difficult for designers to quickly identify a suitable initial structure for complex systems or innovative structures, resulting in low design efficiency and long innovation cycles.

[0003] In recent years, artificial intelligence technologies such as deep learning have demonstrated tremendous potential in various fields. In the field of optical design, some research has attempted to use neural networks to directly infer optical structures or optimize parameters. For example, Paszke et al. demonstrated that deep neural networks (DNNs) can be used to generate initial optical structures, while Côté et al. applied recurrent neural networks (RNNs) to infer lens design sequences. However, these methods often face the following challenges:

[0004] Single optimization objective: Existing methods mostly focus on optimizing a single or a few optical performance indicators (such as RMS (Root Mean Square) spot size), and it is difficult to take into account the complex multi-objective constraints in actual design (such as distortion, MTF (Modulation Transfer Function), size, cost, manufacturability, etc.).

[0005] Weak structure discovery capability: The model mainly interpolates or combines existing design data, making it difficult to proactively discover new and groundbreaking optical structures.

[0006] Insufficient use of prior knowledge: Failure to fully incorporate optical design principles, formulas, and empirical rules to guide the model learning and generation process.

[0007] Large language models (LLMs), such as the GPT (Generative Pre-trained Transformer) series and the Qwen (Tongyi Qianwen) series, offer new solutions to these challenges thanks to their powerful capabilities in text understanding, generation, and zero- and few-shot learning. Through fine-tuning and integration with external knowledge bases (such as retrieval-augmented generation (RAG), LLMs have the potential to understand complex optical design requirements and generate initial optical structures that meet these requirements. Currently, research on applying LLMs, particularly in conjunction with RAG technology, to optical system design is still in its infancy.

[0008] Therefore, developing a method and system that can effectively utilize existing optical design knowledge, understand user needs, and intelligently generate the initial structure of high-quality optical systems is of great significance for improving optical design efficiency and promoting optical technology innovation. Summary of the Invention

[0009] In order to solve the problems in the prior art, the present invention provides an optical system design method and system based on knowledge base retrieval and enhanced large model.

[0010] The technical solutions of the present invention are as follows:

[0011] In a first aspect, the present invention discloses an optical system design method based on knowledge base retrieval and enhanced large model, comprising:

[0012] 1) Obtain and preprocess optical lens design data, vectorize and encode the preprocessed optical lens design data, and build an optical system knowledge base;

[0013] 2) Obtain a parameter list of each surface of the optical system corresponding to the optical system design requirements, construct a design requirement-parameter list data pair, and then construct a training dataset based on the design requirement-parameter list data pair;

[0014] 3) Using the training dataset and a low-rank adaptive method to train the large language model, a large model of optical system design is obtained. The large model of optical system design is then combined with the optical system knowledge base through the knowledge base retrieval enhancement module to obtain a knowledge base retrieval enhancement large model consisting of the knowledge base retrieval enhancement module and the large model of optical system design.

[0015] 4) Obtain the design requirements of the optical system to be designed and input them into the knowledge base retrieval enhancement module for information retrieval to obtain the optical design information most relevant to the design requirements. Then, input the design requirements and optical design information together into the optical system design model to obtain the optical system structural parameters that meet the design requirements of the optical system to be designed, that is, obtain a parameter list of each surface of the optical system.

[0016] In a second aspect, the present invention discloses an optical system design system based on a knowledge base retrieval enhanced large model for implementing the design method, comprising:

[0017] A knowledge base construction module is used to obtain and preprocess optical lens design data, vectorize and encode the preprocessed optical lens design data, and construct an optical system knowledge base;

[0018] A data set construction module is used to obtain a parameter list of each surface of the optical system corresponding to the optical system design requirements, construct a design requirement-parameter list data pair, and then construct a training data set based on the design requirement-parameter list data pair;

[0019] A model building module is used to train the large language model using a training data set and a low-rank adaptive method to obtain a large model of optical system design; the optical system design large model is then combined with the optical system knowledge base through a knowledge base retrieval enhancement module to obtain a knowledge base retrieval enhancement large model consisting of the knowledge base retrieval enhancement module and the optical system design large model;

[0020] The design module is used to obtain the design requirements of the optical system to be designed, and input them into the knowledge base retrieval enhancement module for information retrieval to obtain a number of optical design information most relevant to the design requirements. The design requirements and optical design information are then input together into the optical system design model to obtain the optical system structural parameters that meet the design requirements of the optical system to be designed, that is, to obtain a parameter list of each surface of the optical system.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1) Overcoming the single optimization objective and achieving understanding of complex multi-objective design requirements: To address the pain point of existing methods that find it difficult to take into account multiple constraints such as distortion, MTF, size, and cost, this invention constructs an "optical system design requirements" template that includes multi-dimensional indicators, format requirements, and even optical formulas. By leveraging the powerful text comprehension capabilities of the large language model, it can fully understand and process complex multi-objective design tasks described in natural language, rather than simply optimizing a single RMS spot size, making the generated results closer to actual engineering needs.

[0023] 2) Enhanced structure discovery capabilities promote design innovation: Existing methods primarily interpolate or combine existing data, making it difficult to discover new structures. This invention utilizes augmented retrieval (RAG) technology to retrieve multiple related, but potentially structurally diverse, design examples from a massive knowledge base before generation. A large language model then infers and integrates these examples, rather than simply imitating them. This allows for the emergence of new, groundbreaking, and non-intuitive optical structures, enhancing design innovation.

[0024] 3) Deeply Utilize Prior Knowledge to Improve the Rationality of Generative Solutions: Addressing the pain point of existing methods' insufficient utilization of optical principles and empirical rules, this paper constructs a knowledge base containing a vast amount of optical design data, formulas, and rules. Through RAG technology, prior knowledge most relevant to the current design requirements (such as the applicable scope of specific materials, design paradigms for classical structures, and Abbe invariants) is dynamically injected into the context of the large model during the generation process. This guides the model in generating physically plausible initial structures that conform to engineering practice, significantly improving the usability and quality of the solutions.

[0025] 4) Automation and efficiency improvements, lowering the experience threshold: Combining the aforementioned advantages, this invention models the design thinking and knowledge base of experienced optical engineers. This large model automatically generates high-quality initial optical structures, significantly reducing the time spent on manual conception and trial and error, and significantly improving design efficiency. This allows even inexperienced designers to quickly obtain reasonable, even innovative, design starting points, reducing reliance on experienced engineers and accelerating the overall product iteration cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of the optical system design method based on knowledge base retrieval and enhanced large model of the present invention;

[0027] Figure 2 It is a structural diagram of the optical system design system based on knowledge base retrieval enhanced large model of the present invention;

[0028] Figure 3 is a light path diagram of an optical system generated by using an optical system design method according to an embodiment of the present invention;

[0029] Figure 4 It is the optical path diagram of the optical system generated by the existing method;

[0030] Figure 5 The present invention is Figure 4 The optical path diagram of the optical system generated under the optical system design requirements. DETAILED DESCRIPTION

[0031] The present invention will be further described and illustrated below in conjunction with specific embodiments. The embodiments are merely illustrative of the present disclosure and do not limit its scope. The technical features of the various embodiments of the present invention may be combined accordingly, provided that there is no conflict between them.

[0032] The main purpose of the present invention is to overcome the shortcomings of the existing technology and provide an optical system design method and system based on knowledge base retrieval enhanced large model, aiming to achieve automatic, intelligent and efficient generation of optical system structural parameters.

[0033] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0034] The present invention proposes an optical system design method and system based on knowledge base retrieval enhanced large model. The core of the optical system design method is to achieve the understanding of user optical design requirements and the intelligent generation of initial optical system structure by constructing an optical system knowledge base, fine-tuning a large language model and combining retrieval enhanced generation technology. Figure 1 As shown, the method mainly includes the following steps:

[0035] (1) Construction of optical system knowledge base

[0036] This step aims to build a comprehensive and structured optical system knowledge base in a form that can be efficiently retrieved and utilized by large language models.

[0037] (1.1) Collection and preprocessing of optical lens design data: Collect and organize a large amount of optical lens design data from various sources. For example, it can be extracted and collected from lens library files in ZMX, TXT or XLSX formats that contain a large amount of optical lens design data. A typical lens library file can contain about 2 million optical lens design data. Each optical lens design data contains optical system parameters and / or surface parameters. The optical system parameters include but are not limited to field of view (FOV), F number (f num ), aperture surface index, effective focal length (EFL), RMS spot size (fit), aperture type, aperture value, number of fields of view and wavelength, etc.; surface parameters include but are not limited to the serial number, type (such as STANDARD), remarks (such as OBJ, STOP), curvature radius (mm), thickness (mm), material, refractive index, Abbe number and half-aperture (mm) of each optical surface.

[0038] The collected optical lens design data is then preprocessed, that is, it is cleaned and formatted uniformly. Cleaning refers to screening and deduplication of the optical lens design data. Screening means removing low-quality optical lens design data; low-quality optical lens design data generally refers to optical lens design data with an excessively large RMS spot size. Format unification refers to organizing the cleaned optical lens design data into a structured or semi-structured format (such as TXT, HTML, MarkDown, or JSON file). For example, all cleaned optical lens design data can be aggregated and stored as a structured JSON file to preserve the hierarchical relationship between parameters, or stored as a plain text (TXT) file separated by a specific delimiter (such as "===").

[0039] (1.2) Vectorized encoding of pre-processed optical lens design data and construction of an optical system knowledge base: In order for the large language model to understand and retrieve information in the knowledge base, the pre-processed optical lens design data needs to be converted into a vector representation. The present invention uses an advanced text embedding model, such as BGE-M3, to vectorize the pre-processed optical design data and generate high-dimensional feature vectors for subsequent semantic retrieval. These high-dimensional feature vectors are then stored in a dedicated optical system knowledge base to support subsequent rapid retrieval based on semantic similarity. The optical system knowledge base may, for example, contain high-dimensional feature vectors obtained by vectorizing and encoding more than 2 million pieces of optical lens design data.

[0040] (2) Constructing a training dataset for fine-tuning a large language model

[0041] This step aims to create a dedicated dataset that can train a large language model to understand optical design requirements and generate corresponding structural parameters.

[0042] (2.1) Generation of design requirements-parameter list data pairs: Utilize existing optical design files (such as Zemax file libraries) and extract their detailed optical parameters in batches through automated scripts (such as ZPL macros or Python scripts combined with Zemax API). Simultaneously, simulate the design requests that real users may make and construct optical system design requirements. Each optical system design requirement contains optical system parameters (i.e., target optical system parameters, such as FOV=10.0, f num=2.0, fit = 0.03370, aperture surface position (e.g., aperture is surface.1), etc.), as well as format requirements for the optical system structural parameters output by the large language model (e.g., please give me detailed parameters of every surface in markdown). The design requirements section can also include definitions of the optical system parameters (i.e., definitions of each component within the optical system parameters), requirements for optical system imaging quality at various fields of view, and requirements for optical system imaging quality at different wavelengths at various fields of view. To enhance the large language model's understanding and generation quality, the optical system design requirements can also be embedded with basic optical design formulas (e.g., numerical aperture, Snell's law, Abbe invariant, Lagrange invariant), as well as guiding hints and constraints for the model generation process (e.g., ensuring readability of the output table, prioritizing the use of common optical glass materials, and ensuring that light can pass through all generated surfaces). This can guide the trained large language model to generate more reasonable designs.

[0043] The parameter list section contains detailed parameters of each surface of the optical system that meet the requirements of the optical system design. This is equivalent to the structural parameters of the optical system. It is usually presented in a structured format such as Markdown table, HTML table, JSON text, etc.

[0044] (2.2) Data Pair Formatting: The extracted optical system structural parameters (as parameter lists) are organized according to the format specified in the optical system design requirements (e.g., a Markdown table). Each pair of optical system design requirements and their corresponding parameter lists constitutes a training example. This method generates hundreds or thousands of such data pairs and organizes them into a format suitable for fine-tuning a large language model (e.g., a JSON file). This constitutes the training dataset used for fine-tuning the large language model.

[0045] (3) Efficient fine-tuning of parameters of large language models

[0046] This step aims to adapt the general pre-trained large language model to the field of optical design, so that it has the ability to generate optical system structural parameters according to specific optical system design requirements.

[0047] (3.1) Selection of a basic large language model: Select a pre-trained large language model with strong text understanding and generation capabilities as the basis, such as the Qwen2.5-14B-Instruct model.

[0048] (3.2) Implementation of efficient parameter fine-tuning: Using efficient parameter fine-tuning techniques such as Low Rank Adaptation (LoRA), fine-tune the basic large language model using the training dataset constructed based on the design requirements-parameter list data in step (2). The fine-tuning process is usually performed on a cloud service platform (such as SiliconCloud) with corresponding computing resources or on local computing resources (such as a server with one or more graphics cards).

[0049] During fine-tuning, a series of hyperparameters must be carefully adjusted, including the learning rate, number of epochs, batch size, LoRA technology-specific parameters (such as LoRA rank, LoRA alpha, dropout rate), and the maximum number of tokens in the input sequence. By optimizing these parameters, the fine-tuned large language model accurately understands optical system design requirements and generates optical system structural parameters that meet format and content requirements. After fine-tuning, a large model optimized for the optical design domain is obtained, resulting in an optical system design large model. For example, a model named ft:LoRA / Qwen / Qwen2.5-14B-Instruct is obtained.

[0050] Among them, the cross entropy loss function can be used as the loss function in the process of efficiently fine-tuning the parameters of the large language model.

[0051] (4) Integration and application of knowledge base retrieval enhancement generation (RAG) module:

[0052] This step aims to combine the fine-tuned large language model with the constructed optical system knowledge base to improve the relevance and accuracy of the model generation results.

[0053] (4.1) Configuration of the RAG framework: This is achieved using a mature retrieval enhancement generation framework such as RAGFlow or similar technologies such as DIFY. The optical system knowledge base in step (1.2) is connected to the retrieval module of the RAGFlow framework. At the same time, the service interface of the optical system design large model obtained by fine-tuning in step (3.2) (such as the endpoint address, API key, and model name that comply with the OpenAI API compatibility specification) is configured into the knowledge base retrieval enhancement generation module of the RAGFlow framework, thereby obtaining a knowledge base retrieval enhancement large model consisting of the knowledge base retrieval enhancement module and the optical system design large model. When the knowledge base retrieval enhancement generation module receives input, it can first retrieve one or more of the optical system design examples, surface parameter ranges, or optical system design rules that are most relevant to the optical system design requirements from the optical system knowledge base.

[0054] (4.2) Receiving and processing the design requirements of the optical system to be designed: The knowledge base retrieval enhanced large model receives the design requirements of the optical system to be designed. The format and content of the design requirements of the optical system to be designed should be consistent with the format of the actual optical system design requirements simulated in step (2.1) to ensure that the optical system design large model and the knowledge base retrieval enhanced generation module of the knowledge base retrieval enhanced large model can accurately understand them.

[0055] (4.3) RAG-based knowledge retrieval and context enhancement: When the knowledge base retrieval enhancement generation module receives the optical system design requirements to be designed, its retriever component first vectorizes the optical system design requirements to be designed. Then, it performs a semantic similarity search (e.g., calculating cosine similarity) within the optical system knowledge base constructed in step (1.2) to identify the optical system design cases, surface parameter ranges, optical system design rules, or related formulas that are most relevant to the current optical system design requirements to be designed. In other words, it finds the optical design information. These retrieved information fragments serve as the context of the optical system design requirements to be designed.

[0056] (4.4) Structure Generation Based on Augmented Context: The knowledge base retrieval enhancement generation module inputs the generated optical design information into the optical system design macromodel of the knowledge base retrieval enhancement macromodel. The optical system design macromodel of the knowledge base retrieval enhancement macromodel also receives the design requirements of the optical system to be designed. The design requirements and optical design information form a composite prompt (augmented prompt) with richer content and clearer direction. The macromodel performs reasoning and content generation based on this augmented context, and ultimately outputs optical system structure parameters that meet the design requirements of the optical system to be designed. The output is presented in the format required by the design requirements (e.g., Markdown table, HTML table, JSON structured text, etc.). The optical system structure parameters are a list of parameters for each surface of the optical system, including the surface shape, thickness, material, or refractive index of each surface of the optical system.

[0057] Through the above steps, the present invention realizes a method that can combine large-scale prior knowledge and real-time user needs to intelligently generate optical system structures.

[0058] Implementation Examples

[0059] To verify the effectiveness of the method described in the present invention, experiments were conducted on a server equipped with appropriate computing resources (e.g., a cloud server instance with an NVIDIA GPU).

[0060] 1. Experimental environment and parameter settings:

[0061] Optical system knowledge base: A collection of XLSX files containing approximately 2 million pieces of optical lens design data was used. After preprocessing, the files were vectorized using the BGE-M3 model and stored in the optical system knowledge base.

[0062] Optical system design large model: Qwen2.5-14B-Instruct was selected as the basic large language model for fine-tuning.

[0063] Fine-tuning training dataset: By parsing about 1,000 Zemax design files, about 1,000 pairs of design requirement-parameter list data were automatically generated and formatted as JSONL files.

[0064] Fine-tuning parameters (on local computing resources): The learning rate is set to 0.0001, the number of training iterations (epochs) is 3, the batch size is 8, the LoRA rank is 8, the LoRA alpha value is 32, the dropout rate is 0.05, and the maximum number of tokens is 32768. After successful fine-tuning, the model version ft:LoRA / Qwen / Qwen2.5-14B-Instruct is obtained.

[0065] Knowledge base retrieval enhanced large model: It is built using the RAGFlow framework, and the above-mentioned optical system knowledge base and the fine-tuned optical system design large model are configured. That is, the optical system design large model is combined with the optical system knowledge base through the retrieval enhancement generation technology to obtain a knowledge base retrieval enhanced large model consisting of a knowledge base retrieval enhancement module and the optical system design large model.

[0066] 2. Example of generating optical system structural parameters: The method of the present invention can be used to generate optical system structural parameters according to different design requirements of the optical system to be designed (such as FOV=10, f num =2, the aperture is located in front of the first lens), generate the optical system structure parameters including multiple lenses. The generated optical system structure parameters can be directly imported into optical design software such as Zemax for further image quality evaluation and optimization. Figure 3 As shown, it is a light path diagram of a certain optical system of optical system structural parameters obtained by the optical system design method of the present invention.

[0067] The specifications for the model were: half field of view (HFOV) = 40 degrees, F / # = 4, effective focal length (EFL) = 50 mm, and operating wavelengths of three colors (656 nm, 588 nm, and 486 nm). Furthermore, the design requirements specified the model: "Note that the design must include an aperture and an image plane, and the medium in front of the image plane must be air, with the image plane located at the optimal imaging point."

[0068] As a comparison, Figure 4 This paper demonstrates an optical system generated using existing deep learning methods under the same specifications. This system consists of a relatively simple double-cemented lens system. The optical path diagram shows that light rays of different colors are severely diffused near the image plane, failing to converge into a sharp focus. This indicates significant chromatic aberration and other aberrations, resulting in poor image quality and difficulty meeting design requirements.

[0069] Figure 5 The optical system generated by the method of the present invention under the same specification requirements is shown. Figure 4 In comparison, the system generated by the present invention is a more complex multi-lens system. Its optical path diagram clearly shows that light rays of different colors and fields of view converge into a very small diffuse spot after passing through the system, with a sharp focus. This demonstrates that the system generated by the present invention effectively corrects chromatic aberration and other aberrations, and its optical performance surpasses that of systems generated by existing methods. This comparison fully demonstrates that the present invention, by combining the reasoning power of a large-scale knowledge base and a large language model, can generate more complex, optimized initial optical system structures that better meet high-performance design requirements.

[0070] like Figure 2 As shown, in an embodiment of the present invention, the present invention also provides an optical system design system based on a knowledge base retrieval enhanced large model to implement the design method, including a knowledge base construction module, a data set construction module, a model construction module and a design module.

[0071] The knowledge base construction module is used to obtain and preprocess optical lens design data, vectorize and encode the preprocessed optical lens design data, and construct an optical system knowledge base. The dataset construction module is used to obtain parameter lists for each optical system surface corresponding to the optical system design requirements, construct design requirement-parameter list data pairs, and then construct a training dataset based on the design requirement-parameter list data pairs. The model construction module is used to train a large language model using the training dataset and a low-rank adaptive method to obtain an optical system design large model. The optical system design large model is then combined with the optical system knowledge base through the knowledge base retrieval enhancement module to obtain a knowledge base retrieval enhancement large model consisting of the knowledge base retrieval enhancement module and the optical system design large model. The design module is used to obtain the design requirements of the optical system to be designed, input them into the knowledge base retrieval enhancement module for information retrieval, obtain the optical design information most relevant to the design requirements, and then input the design requirements and optical design information into the optical system design large model to obtain the optical system structural parameters that meet the design requirements of the optical system to be designed, that is, obtain a parameter list for each surface of the optical system.

[0072] The present invention solves the problem of traditional optical design's over-reliance on manual experience through a knowledge-guided intelligent generation mechanism, provides automated initial solution generation capabilities for optical system design, effectively improves optical design efficiency, and is suitable for the innovative design of various imaging optical systems.

[0073] The above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. Persons skilled in the art will readily appreciate that variations and modifications may be made without departing from the scope of the present invention, all of which fall within the scope of protection of the present invention.

Claims

1. An optical system design method based on knowledge base retrieval and enhanced large model, characterized in that: include: 1) Obtain and preprocess optical lens design data, vectorize and encode the preprocessed optical lens design data, and build an optical system knowledge base; 2) Obtain a parameter list of each surface of the optical system corresponding to the optical system design requirements, construct a design requirement-parameter list data pair, and then construct a training dataset based on the design requirement-parameter list data pair; 3) Using the training dataset and a low-rank adaptive method to train the large language model, a large model for optical system design is obtained; Then, the optical system design large model is combined with the optical system knowledge base through the knowledge base retrieval enhancement module to obtain a knowledge base retrieval enhancement large model consisting of the knowledge base retrieval enhancement module and the optical system design large model; 4) Obtain the design requirements of the optical system to be designed and input them into the knowledge base retrieval enhancement module for information retrieval to obtain the optical design information most relevant to the design requirements. Then, input the design requirements and optical design information together into the optical system design model to obtain the optical system structural parameters that meet the design requirements of the optical system to be designed, that is, obtain a parameter list of each surface of the optical system.

2. The optical system design method based on knowledge base retrieval and enhanced large model according to claim 1 is characterized in that: In step 1), optical lens design data is obtained from an existing lens library file, wherein the optical lens design data includes optical system parameters and / or surface parameters, wherein the optical system parameters include field of view angle, F number, aperture surface index, effective focal length, RMS spot size, aperture type, aperture value, field of view number and wavelength, and the surface parameters include the serial number, type, remarks, curvature radius, thickness, material, refractive index, Abbe number and semi-aperture of each optical surface.

3. The optical system design method based on knowledge base retrieval and enhanced large model according to claim 1, characterized in that: The pretreatment in step 1) includes: First, the optical lens design data is screened and deduplicated, and then the format of the screened and deduplicated optical lens design data is unified to complete the preprocessing; The vectorized encoding is: using the BGE-M3 model to vectorize the pre-processed optical lens design data.

4. The optical system design method based on knowledge base retrieval and enhanced large model according to claim 1, characterized in that: In step 2), the optical system design requirements include optical system parameters, definitions of optical system parameters, requirements for optical system imaging quality at various field of view angles, requirements for optical system imaging quality at different wavelengths at various field of view angles, format requirements for optical system structural parameters output by the large language model, optical design formulas, and prompts and constraints for the large language model generation process.

5. The optical system design method based on knowledge base retrieval and enhanced large model according to claim 4 is characterized in that: When constructing a design requirement-parameter list data pair, the parameter list needs to be organized according to the format specified in the corresponding optical system design requirements, and then the parameter list after adjusting the format and the corresponding optical system design requirements form a design requirement-parameter list data pair.

6. The optical system design method based on knowledge base retrieval and enhanced large model according to claim 1, characterized in that: In step 3), the large language model is the Qwen2.5-14B-Instruct model; and the knowledge base retrieval enhancement module is implemented using the RAGFlow framework or the DIFY technology.

7. The optical system design method based on knowledge base retrieval and enhanced large model according to claim 2, characterized in that: In step 4), the design requirements of the optical system to be designed are consistent with the format of the optical system design requirements in the training dataset; The optical design information includes one or more of an optical system design case, a parameter range of a surface parameter, or an optical system design rule.

8. The optical system design method based on knowledge base retrieval and enhanced large model according to claim 1, characterized in that: In step 4), the optical system structural parameters include the surface shape, thickness, material, or refractive index of each surface of the optical system.

9. An optical system design system based on knowledge base retrieval and enhanced large model for implementing the design method of claim 1, characterized in that: include: A knowledge base construction module is used to obtain and preprocess optical lens design data, vectorize and encode the preprocessed optical lens design data, and construct an optical system knowledge base; A data set construction module is used to obtain a parameter list of each surface of the optical system corresponding to the optical system design requirements, construct a design requirement-parameter list data pair, and then construct a training data set based on the design requirement-parameter list data pair; A model building module is used to train the large language model using a training data set and a low-rank adaptive method to obtain a large model of optical system design; the optical system design large model is then combined with the optical system knowledge base through a knowledge base retrieval enhancement module to obtain a knowledge base retrieval enhancement large model consisting of the knowledge base retrieval enhancement module and the optical system design large model; The design module is used to obtain the design requirements of the optical system to be designed, and input them into the knowledge base retrieval enhancement module for information retrieval to obtain a number of optical design information most relevant to the design requirements. The design requirements and optical design information are then input together into the optical system design model to obtain the optical system structural parameters that meet the design requirements of the optical system to be designed, that is, to obtain a parameter list of each surface of the optical system.

Images