Intelligent recommendation method for building color separation prism optical system

Abstract

The invention provides an intelligent recommendation method for building a color separation prism optical system, and belongs to the technical field of intelligent recommendation, and the method comprises the steps: obtaining dominant and implicit granularity features of an optical demand, and building an optical feature matrix after correlation mining; establishing an optical feature priority dynamic attenuation function, hierarchically splitting matrix features through a hierarchical encoder, and mapping the matrix features to a component semantic space in combination with an optical component knowledge graph; inputting a mapping result into a matching model built by an optical component heterogeneous graph neural network, and obtaining a component recommendation combination by taking light splitting precision, component compatibility and working condition adaptability as core optimization targets; and performing parameter correction and optimal value calculation on each combination, and screening a final recommendation combination to complete system establishment. According to the method, through deep excavation of explicit and implicit features, dynamic priority adjustment and multi-constraint component matching, the building precision, stability and adaptability of the color separation prism optical system are improved, meanwhile, the building efficiency is remarkably improved, and personalized requirements under complex scenes are met.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent recommendation, and particularly relates to an intelligent recommendation method for building a dichroic prism optical system. BACKGROUND

[0002] As a core component in the fields of optical imaging, optical communication, and precision detection, the building quality of a dichroic prism optical system directly determines the performance of downstream equipment. Traditional building of a dichroic prism optical system relies on the experience accumulation of technical personnel, and the optical components are selected and combined by manual screening, which has many limitations. On the one hand, it is difficult for manual screening to fully mine the potential demands implied in the optical requirements, and only the components are selected based on the explicit performance indicators, which may lead to insufficient adaptability of the component combination. On the other hand, the priority of the optical characteristics is fixed, and cannot be dynamically adapted to the changes in different stages of the working period of the dichroic prism. Moreover, the component matching process does not fully integrate the multi-dimensional constraints such as light splitting accuracy, component compatibility, and working condition adaptability, which may lead to problems such as low light splitting accuracy, poor stability, and weak working condition adaptability of the built system. At the same time, the manual building is low in efficiency, and cannot meet the high-precision and personalized optical system building requirements in complex scenarios.

[0003] Therefore, the present application provides an intelligent recommendation method for building a dichroic prism optical system. SUMMARY

[0004] The present application provides an intelligent recommendation method for building a dichroic prism optical system to solve the above technical problems.

[0005] The present application provides an intelligent recommendation method for building a dichroic prism optical system, comprising: Step 1: obtaining explicit granularity features and implicit granularity features in optical requirements, and correlatively mining the explicit granularity features and the implicit granularity features to construct an optical feature matrix; Step 2: establishing a dynamic attenuation function of optical feature priority adapting to the core requirements of the dichroic prism, using a hierarchical optical feature encoder to hierarchically split the optical feature matrix according to the feature priority dimension, and combining a pre-constructed optical component knowledge graph to map the split matrix features to an optical component semantic space; Step 3: input the mapping result into the component matching model built by the optical component heterogeneous graph neural network, and optimize the output matching result of the component matching model with the light splitting accuracy, component compatibility and working condition adaptability of the dichroic prism system as the core optimization objectives to obtain the component recommended combination of the dichroic prism optical system, wherein the component recommended combination includes optical processing components and optical building components, and the number of internal components of the optical processing components and the optical building components in different component recommended combinations is not unique, and the performance constraint of the optical adaptation learning framework is fused in the optical component heterogeneous graph neural network; Step 4: label and pre-experiment each component recommended combination to obtain an initial experimental parameter set, and perform maximum and minimum correction on each component recommended combination according to the combination attribute to obtain a modified experimental parameter set pair, and calculate the initial experimental parameter set and the modified experimental parameter set pair of each component recommended combination to determine the optimal value; Step 5: select the component recommended combination corresponding to the final value from all optimal values as the final recommended combination to build the dichroic prism optical system.

[0006] Preferably, the explicit granularity features and the implicit granularity features are associated and mined to construct an optical feature matrix, including: The pre-trained optical demand-feature association topology network is called to directionally mine the potential mapping association between the explicit granularity features and the implicit granularity features corresponding to the optical demand of the dichroic prism, and generate a mining set of each explicit granularity feature and a mining set of each implicit granularity feature , wherein the explicit granularity feature is a concrete optical performance index feature, and the implicit granularity feature is a potential demand dimension feature; The mining set of any one group of explicit granularity features and the mining set of implicit granularity features , analyze the content and set coverage range of the two mining sets and the intersection coincident dimension , quantify the demand transmission strength between the corresponding group of explicit and implicit features to obtain the corresponding demand influence weight value, wherein ; The explicit granularity feature dimension, the implicit granularity feature dimension and the demand influence weight dimension are used as three-dimensional coordinate axes to construct a three-dimensional optical feature matrix special for dichroic prisms, wherein each element in the three-dimensional optical feature matrix uniquely corresponds to the association relationship representation of a group of explicit granularity features and implicit granularity features, and the element carries the demand influence weight value corresponding to the group of associated relationships.

[0007] Preferably, an optical feature priority dynamic decay function suitable for the core demand of the dichroic prism is established, including: Differential attenuation coefficients are set according to three stages of the working cycle of the dichroic prism, and a phased demand timeliness factor is calculated , wherein, are the starting times of the respective stages, are the attenuation coefficients of the respective stages, determined by the time allocation proportion of the working task of the dichroic prism; Based on the demand deviation rate, a real-time demand urgency factor is obtained , and , wherein, is the real-time performance parameter of the kth demand dimension; is the rated standard value of the kth demand dimension; is the urgency gain coefficient; An abnormality-resistant adaptive redundancy factor is determined , wherein, is the current adaptive margin of the kth demand dimension; is the maximum allowed adaptive margin of the kth demand dimension; is the minimum value, and ; Based on the coupling effect of timeliness and urgency represented by the factor interaction term A dynamic attenuation function is constructed , obtaining an intermediate priority weight , and obtaining the optical feature priority weight corresponding to the kth demand dimension at time t , wherein, is the coupling strength coefficient; is the redundancy correction coefficient; is the correction coefficient; is the initial priority weight of the kth demand dimension; The Euclidean distance between adjacent two weights is determined The update period is adaptively adjusted to realize real-time adaptation of the priority weight to the dynamic change of demand.

[0008] Preferably, the split matrix features are mapped to the optical component semantic space in combination with a pre-constructed optical component knowledge graph, including: A hierarchical optical feature encoder with adaptive coding parameter configuration is constructed, the encoder includes three priority-specific coding sub-layers of high, medium and low priority, and the coding parameters of each sub-layer are dynamically configured based on the optical feature priority weight ; The divided high, medium and low priority feature subsets are respectively input into the corresponding coding sub-layers, and the hierarchical coding feature sets are obtained after coding processing , at the same time, the information entropy of each coding feature set is calculated , if , the corresponding coding sub-layer is triggered to re-encode. An optical component knowledge graph with multiple types of associated edges is constructed. The optical component knowledge graph includes four types of nodes: optical processing component nodes, optical construction component nodes, adaptation constraint nodes, and working condition associated nodes, and three types of weighted associated edges: component-component compatibility edges, component-constraint adaptation edges, and component-working condition associated edges. Fusion hierarchy priority matching factor Calculate the basic semantic similarity value ,in, This is the feature vector of the j2th component node in the optical component knowledge graph; Based on multi-constraint similarity pairs The final semantic similarity is obtained by calibration. The node feature vectors are extracted and weighted according to similarity weights to generate a set of semantic mapping vectors for optical components with hierarchical encoding features. This completes the mapping of features to the semantic space of optical components.

[0009] Preferably, the loss function of the heterogeneous graph neural network of the optical components ,in, Cross-entropy loss for component matching tasks; The squared loss for the three types of constraints, To constrain the loss weights, The constraint loss gain coefficient, and The high priority weights corresponding to the spectral accuracy. To ensure component compatibility with corresponding medium priority weights, Adapt the corresponding low-priority weights to the operating conditions; These are constraints on spectral accuracy, component compatibility, and operating condition adaptability.

[0010] Preferably, for each component recommended combination, the maximum and minimum values ​​are adjusted according to the combination attributes to obtain the set of corrected experimental parameters, including: The recommended combination of components for a dichroic prism optical system is analyzed and classified by attributes to determine performance attributes, structural adaptability attributes, operating condition compatibility attributes, and characteristic parameters. Real-time priority weighting based on optical features Determine the priority order of various attributes and the dynamic boundary constraints; For each component, a recommended combination is made, and targeted maximum and minimum corrections are performed according to the attribute type. The correction parameters for each type of attribute are then summarized to form a set of correction experiment parameters.

[0011] Preferably, the optimal values ​​are determined by calculating the initial experimental parameter set and the corrected experimental parameter set for each component's recommended combination, including: Calculate the comprehensive base score for each component recommendation combination based on the corresponding initial experimental parameter set, maximum correction experimental parameter set, and minimum correction experimental parameter set; The basic comprehensive score for different sets of experimental parameters is corrected to obtain the corrected comprehensive score, and the consistency coefficient of the recommended combination of each component is calculated. If the consistency coefficient is greater than the preset coefficient, the corrected comprehensive score is weighted and calculated to obtain the optimal value. Otherwise, the score corresponding to the parameter set with the highest score after correction is retained first, and the items below the indicator threshold are fine-tuned by combining the individual scores of each evaluation indicator in the corresponding parameter set. After fine-tuning, the comprehensive score is recalculated as the optimal value.

[0012] Preferably, the optical processing component is used to filter and shape the incident light to obtain the color beam to be separated, and the optical assembly component is used to perform beam separation processing on the color beam to be separated.

[0013] Compared with the prior art, the beneficial effects of this application are: By inputting the optical feature mapping results into a component matching model that incorporates performance constraints, and optimizing the core objectives of spectral accuracy, component compatibility, and operational adaptability, a recommended combination of components is obtained. After preliminary experiments, parameter correction, and optimal value calculation, the final combination is selected and the system is built. This achieves intelligent matching and precise optimization of component combinations, while also taking into account the optimal performance and basic stability of the system through parameter correction. At the same time, the flexibility in the number of internal components adapts to different complexity requirements, effectively improving the construction quality and efficiency of the dichroic prism optical system.

[0014] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an intelligent recommendation method for constructing a dichroic prism optical system according to an embodiment of the present invention. Detailed Implementation

[0018] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0019] This invention provides an intelligent recommendation method for constructing a dichroic prism optical system, such as... Figure 1 As shown, it includes: Step 1: Obtain the explicit and implicit granular features in the optical requirements, and perform correlation mining on the explicit and implicit granular features to construct an optical feature matrix; Step 2: Establish a dynamic attenuation function for optical feature priority that adapts to the core requirements of the dichroic prism. Use a hierarchical optical feature encoder to split the optical feature matrix into layers according to the feature priority dimension. Combine the pre-built optical component knowledge graph to map the split matrix features to the optical component semantic space. Step 3: Input the mapping results into the component matching model built by the heterogeneous graph neural network of optical components, and optimize the output matching results of the component matching model with the core optimization objectives of the dichroic prism system's spectral accuracy, component compatibility, and working condition adaptability to obtain the recommended component combination of the dichroic prism optical system. The recommended component combination includes optical processing components and optical construction components, and the number of internal components of the optical processing components and optical construction components in different recommended component combinations is not unique. The heterogeneous graph neural network of optical components incorporates the performance constraints of the optical adaptation learning framework. Step 4: Mark and conduct preliminary experiments for each recommended combination of components to obtain an initial experimental parameter set. Then, perform maximum and minimum corrections on each recommended combination of components according to the combination attributes to obtain a corrected experimental parameter set pair. Finally, calculate the initial experimental parameter set and the corrected experimental parameter set pair for each recommended combination of components to determine the optimal value. Step 5: Select the component recommendation combination corresponding to the final value from all the optimal values ​​as the final recommended combination, and build the dichroic prism optical system.

[0020] Preferably, the optical processing component is used to filter and shape the incident light to obtain the color beam to be separated, and the optical assembly component is used to perform beam separation processing on the color beam to be separated.

[0021] In this embodiment, explicit granularity characteristics refer to the concrete and directly quantifiable optical performance indicators in the optical requirements of dichroic prisms. For example, if the user requires the dichroic prism to have a spectral band range of 400nm-700nm, a transmittance of ≥95%, and a spectral efficiency of ≥90%, the performance indicators that can be directly detected by instruments such as a spectrometer and a power meter are explicit granularity characteristics.

[0022] In this embodiment, implicit granularity features refer to the requirement dimensions of the dichroic prism that are not explicitly mentioned by the user but have a potential impact on the system construction effect. For example, if the user does not explicitly state that the dichroic prism is used for optical inspection equipment in outdoor environments, its high temperature environment (above 40°C) adaptation requirements and long-term operational stability requirements of no performance degradation after 24 hours of continuous operation are implicit granularity features.

[0023] In this embodiment, the optical feature matrix refers to the exclusive matrix of the dichroic prism constructed with the explicit granularity feature dimension, the implicit granularity feature dimension, and the demand influence weight dimension as three-dimensional coordinate axes.

[0024] In this embodiment, the optical feature priority dynamic decay function refers to a mathematical function that dynamically adjusts the priority weights of each optical feature, taking into account factors such as timeliness, urgency of demand, and redundancy of anomaly adaptation at different stages of the dichroic prism's working cycle, so that the priority weights can adapt to the dynamic changes in demand in real time.

[0025] In this embodiment, the optical component semantic space refers to an abstract feature space. The hierarchical features after splitting the optical feature matrix are mapped to vectors in this space. The distance or similarity between vectors characterizes the semantic association between optical features and optical components. In this embodiment, the optical component heterogeneous graph neural network refers to a neural network designed for heterogeneous graph data (optical component knowledge graph). It integrates the performance constraints of the optical adaptation learning framework to learn the relationships between optical components and the matching rules with optical features, providing model support for component matching.

[0026] In this embodiment, the mapping result is... .

[0027] In this embodiment, the component matching model is a model built on a heterogeneous graph neural network, which is used to match optical processing and build components according to the mapping results. For example, after inputting the mapping results, the model outputs combinations such as filter F1 + prism P1 and filter F2 + prism P2.

[0028] In this embodiment, the beam splitting accuracy is the degree to which the dichroic prism system separates incident light according to a preset wavelength band; the component compatibility is the degree of fit between the optical processing and assembly components in terms of structural size, interface specifications, etc.; and the working condition adaptability is the degree of performance stability of the component assembly in the target working environment.

[0029] In this embodiment, the core optimization objective is to prioritize the performance indicators during component matching, namely, spectral accuracy, component compatibility, and operating condition adaptability. These indicators determine the core performance of the dichroic prism system.

[0030] The recommended component combination is a combination of optical processing and building components output after model optimization, such as filter F1 + shaping lens L1 + prism P1 + beam splitter S1. Different combinations have different numbers of internal components.

[0031] Optical processing components are used to filter and shape incident light. For example, they consist of a filter and a shaping lens. The filter filters out light that is not in the target wavelength range, and the shaping lens adjusts the incident light into parallel light to obtain the beam to be separated into colors. The filtering element is selected according to the target wavelength range, and the shaping element is selected according to the beam requirements. The number of components is determined by the complexity of the requirements.

[0032] Optical assembly components are used to split the pre-processed light beam to be separated into different colors. For example, they consist of a prism body and a beam splitting film. After the light beam to be separated enters, it is separated into red, green and blue monochromatic light. The prism body is selected according to the beam splitting band, and the beam splitting film is selected according to the band separation requirements. The number of components is determined by the beam splitting complexity.

[0033] The number of internal components refers to the number of sub-components contained in each optical processing / assembly component. For example, an optical processing component may contain 2 components (filter + shaping lens) or 1 integrated module. The number is not unique. Depending on the integration and complexity of the optical requirements, different numbers of components are selected to balance performance and structural compactness.

[0034] The performance constraints of the optical adaptation learning framework are related to the adaptation of optical systems incorporating heterogeneous graph neural networks, such as the deviation threshold of beam splitting accuracy and the fitting requirements of component compatibility.

[0035] In this embodiment, the recommended component combinations are labeled by assigning a unique identifier to each recommended combination to distinguish it from the pre-experimental data. For example, combination 1 is labeled as Z1 and combination 2 is labeled as Z2.

[0036] Preliminary experiments are initial experiments conducted on the recommended combinations of labeled components to obtain basic performance parameters. For example, spectral testing is performed on combination Z1 to obtain data such as spectral bands and transmittance. By building a small experimental device, components are assembled according to the recommended combinations, and performance parameters are collected and recorded using optical detection equipment.

[0037] The initial experimental parameter set is a set of basic performance parameters for the recommended combination of components obtained after the pre-experiment. For example, the initial parameter set of combination Z1 includes a beam splitting band of 405-695nm, a transmittance of 92%, and an installation size of ±0.05mm. Specifically, it is a structured set of performance, structural, and operating condition parameters collected in the pre-experiment.

[0038] Combination attributes are the characteristic parameters of recommended component combinations, which are divided into performance-related attributes (such as spectral band and transmittance), structural adaptation attributes (such as installation size and interface type), and operating condition compatibility attributes (such as temperature and humidity tolerance range).

[0039] Maximizing correction involves adjusting attribute parameters in a direction that improves the performance of the component combination. For example, the transmittance is corrected from 92% to 94.88%, which is close to the upper limit of 98% of the requirement. Based on the dynamic boundary constraints of the attribute (determined by priority weight), the parameters are adjusted in the direction of optimal performance to ensure that they do not exceed a reasonable range.

[0040] Minimal correction involves adjusting attribute parameters in a direction that meets the basic operational requirements of the component assembly. For example, correcting the installation dimension tolerance from ±0.07mm to the basic threshold of ±0.05mm is based on the basic operational threshold of the attribute, adjusting the parameters in a direction that meets the minimum requirements, and ensuring stable system operation.

[0041] A set of corrected experimental parameters is a collection of parameters after maximum and minimum corrections. For example, the maximum parameter set of combination Z1 is spectral band 400-700nm with transmittance of 94.88%, and the minimum parameter set is spectral band 405-695nm with transmittance of 92%. Together, they constitute a set of parameters.

[0042] The optimal value is the score corresponding to the optimal parameters obtained after calculating the combined score of the initial and modified parameter sets.

[0043] In this embodiment, the incident light is the original light entering the dichroic prism system, such as natural light or laser light in industrial testing equipment. That is, the incident light that meets the requirements is provided by the light source device and connected to the system input terminal.

[0044] Filtering is the operation of optical processing components to filter out non-target wavelength light from incident light. For example, to filter infrared light and retain visible light in the 400-700nm range, a filter element (such as a filter sheet) of the corresponding wavelength band is selected and placed in the path of the incident light to complete the filtering.

[0045] Shaping preprocessing is the operation of optical processing components to adjust the shape of the incident light beam, such as adjusting divergent light into parallel light; the means of implementation is to select shaping elements (such as lenses) and place them according to the optical path design to adjust the beam shape.

[0046] The beam to be separated is a beam that has undergone filtering and shaping preprocessing and is waiting to be separated. For example, the processed 400-700nm parallel visible light is filtered and shaped by the optical processing components to obtain a beam that meets the requirements for beam separation.

[0047] Spectroscopic processing involves using optical components to separate a beam of light into multiple monochromatic beams, such as red light (620-700nm), green light (500-560nm), and blue light (400-480nm). Spectroscopic separation is achieved through the cooperation of a prism and a spectroscopic film, utilizing the principles of light refraction and reflection.

[0048] In this embodiment, for example, the assembly includes at least a first dichroic prism, a second dichroic prism, and a third dichroic prism. The first face of the first dichroic prism is vertically positioned, and an air gap is maintained between the third face of the first dichroic prism and the second dichroic prism. The second face of the second dichroic prism is in contact with the third dichroic prism. The first, second, and third dichroic prisms perform beam splitting processing on the light beam to be separated, outputting beams of different colors from the first, second, and third dichroic prisms. The first face of the first dichroic prism is used to input the light beam to be separated, and the third face of the first dichroic prism is used to reflect the blue light in the light beam to be separated back to the first face of the first dichroic prism. The blue light is reflected back onto the first face of the first dichroic prism, and the reflected blue light is transmitted vertically through the second face of the first dichroic prism and exits the prism.

[0049] The reflection condition for total internal reflection of blue light by the first face of the first dichroic prism is: the first total internal reflection angle is n times the angle between the first face and the third face of the first dichroic prism, where n is a natural number greater than 1.

[0050] The beam of light to be separated is transmitted through the third surface of the first dichroic prism to obtain red light in the beam of light to be separated. The first surface of the second dichroic prism is used to receive the red light, and the second surface of the second dichroic prism is used to reflect the red light back to the first surface of the second dichroic prism. The red light undergoes total internal reflection on the first surface of the second dichroic prism, and the reflected red light is transmitted perpendicularly out of the prism through the third surface of the second dichroic prism.

[0051] The reflection condition for total internal reflection of red light by the first face of the second dichroic prism is: the second total internal reflection angle is determined based on the included angle of the first prism and the included angle of the second prism, where the included angle of the second prism is the included angle between the first face and the second face of the second dichroic prism.

[0052] In this embodiment, an internal component refers to the smallest unit in an optical processing component or optical assembly that can independently perform a specific function. An integrated module is considered as a single internal component (its internal sub-components are not counted separately in the number of components). For example, the filter and shaping lens of an optical processing component are considered as two internal components, and the integrated filter and shaping module is considered as one internal component.

[0053] The beneficial effects of the above technical solution are as follows: by inputting the optical feature mapping results into the component matching model with performance constraints, the recommended combination of components is obtained with the core optimization objectives of spectral accuracy, component compatibility, and working condition adaptability. Then, through pre-experimentation, parameter correction, and optimal value calculation, the final combination is selected and the system is built. This not only achieves intelligent matching and precise optimization of component combinations, but also takes into account the optimal performance and basic stability of the system through parameter correction. At the same time, the flexibility of the number of internal components can adapt to the needs of different complexities, effectively improving the construction quality and efficiency of the dichroic prism optical system.

[0054] This invention provides an intelligent recommendation method for constructing a dichroic prism optical system, which involves correlation mining of explicit and implicit granular features to construct an optical feature matrix, including: The pre-trained optical requirement-feature association topology network is invoked to perform targeted mining of the potential mapping associations between the explicit and implicit granular features corresponding to the optical requirements of the dichroic prism, and to generate a mining set for each explicit granular feature. And the mining set for generating each latent granular feature. The explicit granularity features are concrete optical performance index features, and the implicit granularity features are potential demand dimension features. Mining set of any set of explicit granular features Mining set with latent granular features Analyze the content union coverage of the two sets of mining sets. and overlapping dimensions The strength of demand transmission between the explicit and implicit features of corresponding groups is quantified to obtain the corresponding demand influence weight values, where, ; Using the explicit granularity feature dimension, the implicit granularity feature dimension, and the demand influence weight dimension as three-dimensional coordinate axes, a three-dimensional optical feature matrix exclusive to the color separation prism is constructed. Each element in the three-dimensional optical feature matrix uniquely corresponds to a set of relationships between explicit and implicit granular features, and the element carries the demand influence weight value corresponding to the set of relationships.

[0055] In this embodiment, the optical requirement-feature association topology network refers to a graph neural network model pre-trained based on a large amount of optical requirement and feature association data for dichroic prisms. It is used to directionally mine the potential mapping relationship between explicit and implicit granular features. Network nodes represent optical requirement items and feature items, and edges represent the association strength between them. This topology network is constructed using a graph convolutional neural network (GCN). Training data comes from technical manuals, engineering experimental data, and user feedback cases from optical equipment manufacturers. During network training, the feature association prediction accuracy is optimized. Training stops when the accuracy reaches 90% or higher, and network parameters are saved for later use. The network includes 1000 dichroic prism application cases, each covering 5-8 explicit granular features (such as spectral band, transmittance, and spectral efficiency) and 3-5 implicit granular features (such as high-temperature adaptability and long-term stability). The feature dimension is uniformly 20 dimensions, and the label represents the association strength (0-1) between explicit and implicit features.

[0056] In this embodiment, for and The number of elements in the union of elements with the characteristic of existence; for The number of elements present in it; for The number of elements present in it; for and The number of elements in the intersection of the existence characteristics.

[0057] In this embodiment, the calculation method for the demand impact weight value is as follows: The weighted summation algorithm is used to calculate the weight value of the demand influence. The weight coefficient is determined according to the application scenario of the dichroic prism. In the high-precision optical inspection scenario, the weight coefficient of the intersection and overlap dimension is set to 0.6, and in the general scenario, it is set to 0.5 to ensure that the weight value can accurately reflect the demand transmission intensity.

[0058] In this embodiment, assuming there are 3 dominant features (A1, A2, A3) and 2 latent features (B1, B2), with demand influence weights of W11=1.08, W12=0.92, W21=0.85, W22=1.15, W31=0.78, and W32=0.96 respectively, the elements (A1, B1, 1.08), (A1, B2, 0.92), etc. in the three-dimensional optical feature matrix correspond to the association relationships and weights of each group. Specifically, a matrix construction tool (such as Python's NumPy library) is used to construct an m×n×1 three-dimensional matrix with the number of dominant features m as the x-axis dimension, the number of latent features n as the y-axis dimension, and the demand influence weights as the z-axis values. Each element in the matrix stores the demand influence weight value of the corresponding group, which is convenient for subsequent hierarchical splitting processing.

[0059] The beneficial effects of the above technical solution are: by accurately mining the potential associations of explicit and implicit features through pre-trained associated topology networks, and by combining the union coverage and intersection overlap dimensions to quantify the intensity of demand transmission, the constructed three-dimensional optical feature matrix can comprehensively and accurately characterize the feature associations of optical needs, providing accurate feature data support for subsequent feature layering and component matching, and effectively avoiding the problem of insufficient component matching due to insufficient feature association mining.

[0060] This invention provides an intelligent recommendation method for constructing a dichroic prism optical system, establishing a dynamic attenuation function for optical feature priority that adapts to the core requirements of the dichroic prism, including: Differentiated attenuation coefficients are set according to the three stages of the working cycle of the dichroic prism, and the timeliness factor of the staged demand is calculated. ,in, These represent the start times of each stage. These are the attenuation coefficients for each stage, determined by the time allocation ratio of the dichroic prism's working tasks. Based on dynamic adjustment of the demand deviation rate, the real-time demand urgency factor is obtained. ,and ,in, For the k-th demand dimension, the real-time performance parameter is denoted as . This represents the standard value for the k-th demand dimension; This is the urgency gain coefficient; Determine the anomaly adaptation redundancy factor ,in, This represents the current adaptation margin for the k-th demand dimension; This represents the maximum allowable adaptation margin for the k-th requirement dimension. It is a local minimum, and ; Based on factor interaction terms The coupling effect of the timeliness and urgency of the representation on the constructed dynamic decay function To obtain the intermediate priority weight And obtain the optical feature priority weights corresponding to the k-th demand dimension at time t. ,in, This is the coupling strength coefficient; This is a redundancy correction factor; This is a correction factor; Let be the initial priority weight for the k-th demand dimension; Determine the Euclidean distance between two adjacent weights The update cycle is adaptively adjusted to achieve real-time adaptation of priority weights to dynamic changes in requirements.

[0061] In this embodiment, the three-stage working cycle of the dichroic prism refers to the complete working process of the dichroic prism from start-up to shutdown, which is divided into three stages: the start-up stage, the steady-state stage, and the shutdown preparation stage. The working state and key requirements of each stage are different. The start-up stage is the stage from equipment startup to performance stabilization, the steady-state stage is the stage of stable performance operation, and the shutdown preparation stage is the transition stage before the equipment is about to shut down. For example, if a dichroic prism is used in industrial testing equipment with a working cycle of 12 hours, the start-up stage is 0-1 hours (t0=0, t1=1), the steady-state stage is 1-11 hours (t1=1, t2=11), and the shutdown preparation stage is 11-12 hours (t2=11). Specifically, the working time data is collected by the control system of the equipment where the dichroic prism is located. Based on the working principle and performance characteristics of the equipment, the time range of the three stages is divided. The specific values ​​of t0, t1, and t2 can be adjusted according to the actual application scenario.

[0062] The differentiated attenuation coefficient refers to the attenuation coefficient set for each of the three stages of the dichroic prism's working cycle. It is used to adjust the attenuation rate of the time-sensitive factor required in each stage. The attenuation coefficient is determined by the time allocation ratio of the dichroic prism's working tasks; the higher the time proportion, the smaller the attenuation coefficient, and the slower the attenuation of the time-sensitive factor. For example, in the aforementioned 12-hour working cycle, the startup period accounts for 8.3% (1 / 12), and the attenuation coefficient is set accordingly. =0.8; steady-state time accounts for 83.3% (10 / 12), set =0.2; the downtime preparation period accounts for 8.3% (1 / 12), set =0.7, specifically: The time allocation ratio of each stage within the working cycle of the color-separating prism is statistically analyzed. A linear mapping algorithm is used to convert the time proportion into an attenuation coefficient, with a mapping range of 0.1-0.9. This ensures that the timeliness factor of each stage accurately reflects the time attenuation characteristics of demand. Through 100 sets of color-separating prism operation experiments with different working cycles (2 hours-24 hours), the correlation between the time proportion of each stage and the timeliness attenuation law of demand is statistically analyzed. A linear mapping formula is obtained through fitting, ensuring that the attenuation coefficient accurately reflects the demand attenuation characteristics of different stages. For example, , where x is the time percentage of each stage in the working cycle of the dichroic prism.

[0063] In this embodiment, the performance parameters of the k-th demand dimension are collected in real time using an optical detection device. Retrieve the preset rated standard value Calculate the deviation rate, and substitute it into the formula to calculate the urgency factor. The value ranges from 1 to 3, and is adjusted according to the importance of the requirement. For high-precision optical inspection scenarios (such as semiconductor inspection), the value is [value missing]. =2.5−3, for general scenarios (such as ordinary imaging devices) take =1−1.5, taken in industrial testing scenarios =1.5−2.5, verified through 50 sets of simulation experiments with different urgency levels of demand. The value is positively correlated with the accuracy of the demand response. When the value is 2.5, the response speed of component matching for urgent needs is improved by 40%.

[0064] In this embodiment, the current adaptation margin of the k-th demand dimension is collected by a sensor. Retrieve the preset maximum allowable adaptation margin Substitute into the formula to calculate the redundancy factor. To avoid the denominator being zero.

[0065] In this embodiment, The value range is 0.3-0.7. The value range is 0.2-0.4. The value range is 0.9-1.1 to ensure that the function can accurately reflect the influence of each factor on the priority.

[0066] In this embodiment, the Euclidean distance formula is used to calculate the distance between the two, and the update period is adaptively adjusted according to the distance. For example, when At that time, the update cycle is shortened to 0.5 hours; when The update cycle remains at 1 hour.

[0067] The beneficial effects of the above technical solution are: by dividing the working cycle of the color separation prism into three stages and introducing multi-dimensional influencing factors to construct a dynamic decay function, the priority weight of optical features can be dynamically adjusted in real time. The weight can accurately respond to the time changes, urgency and anti-anomaly capabilities of the requirements, avoid the adaptation lag problem caused by fixed priorities, provide a priority basis that fits the actual needs for feature layering and splitting, and improve the pertinence and rationality of subsequent component matching.

[0068] This invention provides an intelligent recommendation method for constructing a dichroic prism optical system. It combines a pre-built optical component knowledge graph to map the split matrix features to the semantic space of the optical components, including: A hierarchical optical feature encoder with adaptive encoding parameter configuration is constructed. The encoder contains three dedicated encoding sub-layers with high, medium, and low priorities, and the encoding parameters of each sub-layer are based on the priority weights of optical features. Dynamic configuration; The high, medium, and low priority feature subsets are input into their respective coding sub-layers, and after coding processing, hierarchical coded feature sets are obtained. Simultaneously, the information entropy of each encoded feature set is calculated. ,like If so, the corresponding encoding sublayer will be re-encoded; An optical component knowledge graph with multiple types of associated edges is constructed. The optical component knowledge graph includes four types of nodes: optical processing component nodes, optical construction component nodes, adaptation constraint nodes, and working condition associated nodes, and three types of weighted associated edges: component-component compatibility edges, component-constraint adaptation edges, and component-working condition associated edges. Fusion hierarchy priority matching factor Calculate the basic semantic similarity value ,in, This is the feature vector of the j2th component node in the optical component knowledge graph; Based on multi-constraint similarity pairs The final semantic similarity is obtained by calibration. The node feature vectors are extracted and weighted according to similarity weights to generate a set of semantic mapping vectors for optical components with hierarchical encoding features. This completes the mapping of features to the semantic space of optical components.

[0069] In this embodiment, the real-time priority weights output by the optical feature priority dynamic attenuation function are received. and in accordance with The threshold range is used to divide the optical feature matrix features into high, medium, and low priority feature subsets; The dimensionality and information density of each priority feature subset are dynamically configured with encoding parameters: High-priority feature encoding sublayer: Configured with an adaptive 1×n convolutional structure, where n is the kernel length, determined by the number of high-priority feature dimensions. Sure, An attention mechanism is used to enhance the core feature dimension, and ≥92% of the original feature information entropy is retained during the encoding process; This is the floor symbol.

[0070] Medium-priority feature encoding sublayer: Configure a dynamic fully connected encoding structure, the number of hidden layer nodes is adaptively adjusted from input dimension × 0.8 to input dimension × 1.2, and BatchNorm normalization is used to suppress feature redundancy; Low-priority feature encoding sublayer: Configured with an adaptive pooling ratio max pooling structure, its pooling ratio By combining L1 regularization to eliminate invalid feature dimensions, ≤65% of the core feature dimensions are ultimately retained. Indicates all The maximum value in.

[0071] In this embodiment, It is calibrated by the requirement of integrity of optical feature information.

[0072] In this embodiment, the optical processing component and the optical assembly component node carry attribute information such as component model, performance parameters, and compatibility specifications. Adaptive constraint nodes: Bearing constraints such as spectral accuracy threshold and installation dimensional tolerance; Operating condition related nodes: operating condition parameters such as temperature and humidity adaptation range and vibration tolerance threshold; Component-component compatible edge: weight It is quantified by both historical pairing success rate and compatibility lifespan. .

[0073] Component - Constraint Adaptation Edge: Weight The fitting measurement of components that satisfy constraints is quantified. Component-Work Condition Related Edges: Weight It is quantified by the stability test results of the component under the target operating conditions; Set update trigger threshold If the weight change of any associated edge If a new component / condition node is added, the graph node features and edge weights will be automatically updated.

[0074] In this embodiment, i2=1, at this time, i2=2, at this time, i2=3, at this time, .

[0075] In this embodiment, adaptive coding parameter configuration refers to the hierarchical optical feature encoder dynamically adjusting coding parameters, such as convolutional kernel size, number of fully connected layer nodes, and pooling ratio, for each priority coding sub-layer based on the priority weight of the corresponding optical feature, to adapt to the information density and dimensionality of features with different priorities. For example, the weight of high-priority features... (≥0.6), number of feature dimensions =25, then the kernel length n= =5, configured with a 1×5 convolutional structure; medium priority feature weights =0.4 (0.3 < 0.4 < 0.6), input dimension = 20, hidden layer node number adjusted to 16-24; low priority feature weights =0.2 (≤0.3), the maximum weight is 0.7, then the pooling ratio r = 1 - 0.2 / 0.7.

[0076] In this embodiment, the hierarchical coded feature set F refers to the feature set obtained by splitting the optical feature matrix into three feature subsets (high, medium, and low) according to priority and then encoding them through the corresponding coded sub-layers, including high-priority coded features. Medium priority coding features Low-priority coding features Each subset is a set of feature vectors. For example, after splitting the optical feature matrix, the high-priority feature subset contains features such as spectral accuracy and transmittance, which are encoded to obtain... =[0.85,0.92,0.78]; The medium-priority feature subset includes features such as component compatibility and installation size, which are encoded as follows: =[0.63,0.58,0.71]; The low-priority feature subset includes features such as working condition adaptability and appearance dimensions, which are encoded as follows: =[0.42,0.39,0.45].

[0077] In this embodiment, information entropy Refers to the feature set used to measure hierarchical coding (i2=1,2,3 correspond to high, medium, and low priority, respectively) This is an indicator of information richness. A higher information entropy indicates richer information in the feature set and better encoding performance; conversely, a lower information entropy indicates more missing information, requiring re-encoding. For example, Information entropy =2.8, which is greater than the threshold of 2.5, so the encoding is valid; Information entropy =2.3, less than the threshold of 2.5, triggering re-encoding of the low-priority encoding sub-layer. Information entropy calculation formula. ,in, The probability distribution of each element in the feature vector is used to determine the coding effect by calculating the information entropy of each hierarchical coding feature set, ensuring that the coded features can fully reflect the original feature information.

[0078] Hierarchical priority matching factor This refers to the weighting factor corresponding to the hierarchical encoding feature set, used to adjust the importance of features of different priorities in semantic similarity calculation. High-priority features have larger matching factors, while low-priority features have smaller matching factors. For example, the weighting factor corresponding to a high-priority feature... =0.6, corresponding to medium priority features =0.3, corresponding to low-priority features =0.1, ensuring that high-priority features dominate semantic matching. Specifically, the hierarchical priority matching factor is determined based on the priority weights of optical features. It is proportional to the average weight of the corresponding priority, and This ensures that the weight allocation of features at each level is reasonable.

[0079] In this embodiment, the semantic similarity base value is calibrated with multiple constraints (such as combining component compatibility, constraint adaptability and other constraints) to obtain the final semantic similarity. The corresponding component node feature vectors are extracted, and they are weighted and fused according to the similarity weight to generate a hierarchical semantic mapping vector set M, thus completing the feature mapping.

[0080] The beneficial effects of the above technical solution are: by constructing a hierarchical optical feature encoder with adaptive parameter configuration, the encoding effect of features with different priorities is ensured; by constructing an optical component knowledge graph with multiple types of nodes and associated edges, the component association information is enriched; and by semantic similarity calculation and weighted fusion, the accurate mapping of features to the component semantic space is achieved, providing semantically closely related input data for subsequent component matching models and improving the accuracy of component matching.

[0081] This invention provides an intelligent recommendation method for constructing a dichroic prism optical system, wherein the loss function of the heterogeneous graph neural network of the optical components... ,in, Cross-entropy loss for component matching tasks; The squared loss for the three types of constraints, To constrain the loss weights, The constraint loss gain coefficient, and The high priority weights corresponding to the spectral accuracy. To ensure component compatibility with corresponding medium priority weights, Adapt the corresponding low-priority weights to the operating conditions; These are constraints on spectral accuracy, component compatibility, and operating condition adaptability.

[0082] In this embodiment, ,in, The components output by the network are combined to split the optical bands. This refers to the standard value for the spectral band in optical requirements, and Where 0.7 is the benchmark value for high priority weight; In this embodiment, ,in, Weights for compatible edges between components; For component installation dimensional deviations; Allowable deviations in installation dimensions, ; In this embodiment, ,in, This refers to the drift in component performance parameters caused by changes in operating conditions. These are the standard values ​​of performance parameters under reference operating conditions; .

[0083] In this embodiment, the heterogeneous graph neural network for optical components includes an input layer, three stacked hidden layers, and an output layer. Each layer achieves deep coupling between performance constraints and network computation through a dedicated constraint fusion module. Input layer: Configure constraint filtering gating units, which concatenate component node features with three types of constraint parameters and input them into the gating function. ,in, For Sigmoid activation function, To constrain the gate weight matrix, For the bias term, only the gated output is retained. The component node features are then used in subsequent calculations, and invalid nodes that do not initially meet the core constraints are removed. Hidden Layer: A constrained attention mechanism is introduced. The node association weight matrix A of each hidden layer is generated by coupling feature similarity and constraint co-satisfaction. The matrix elements are defined as follows: ,in, The feature cosine similarity between nodes i3 and j3; The constraint coordination satisfaction degree between the two nodes; Attention allocation is used to strengthen the association strength of nodes that satisfy the constraints; Output layer: Construct a constraint-enhanced loss function that incorporates constraint satisfaction into the model optimization objective. The loss function is defined as L, and the matching accuracy and constraint satisfaction are simultaneously optimized through backpropagation.

[0084] In this embodiment, the model outputs a matching probability of [0.92, 0.05, 0.03] for a certain component combination, and the true label is [1, 0, 0]. The cross-entropy loss is then calculated as follows: The loss value is small, and the matching effect is good. ,in, This is a real label; The model outputs the matching probability, and this loss is used as one of the core objectives of model optimization. The model parameters are adjusted through backpropagation.

[0085] In this embodiment, the constraint squared loss is calculated based on the degree of satisfaction of the three types of constraints C1, C2, and C3, combined with the corresponding constraint loss weights w01, w02, and w03, and substituted into the squared loss formula. The weights are equal to the priority weights corresponding to the spectroscopic accuracy, component compatibility, and operating condition adaptability, respectively, to ensure that the constraint loss can reflect the importance of different constraints. Specifically, the constraint loss weights for spectroscopic accuracy are w01=0.6 and C1=0.95; for component compatibility, w02=0.3 and C2=0.88; and for operating condition adaptability, w03=0.1 and C3=0.90.

[0086] In this embodiment, The value ranges from 0.3 to 0.8, depending on the application scenario of the dichroic prism. For example, in high-precision scenarios, Choose a larger value, 0.7-0.8, to ensure the model prioritizes meeting the constraints; in general scenarios, A value of 0.3-0.5 is used to balance matching accuracy and constraint satisfaction.

[0087] In this embodiment, the compatibility edge weights between components in the component knowledge graph are obtained. Calculate the component installation size deviation, substitute it into the formula to calculate C2, and then compare it with the threshold. Compare and ensure that the compatibility of the component combinations meets the requirements.

[0088] In this embodiment, the performance drift of the component combination is obtained through a simulated operating condition change experiment. Retrieve standard values ​​of performance parameters under reference operating conditions Substitute into the formula to calculate C3, and compare with the threshold. Comparisons ensure that the component combination can adapt to the target operating conditions.

[0089] The beneficial effects of the above technical solution are as follows: by constructing a total loss function that integrates matching loss and constraint loss, the model optimization objectives are incorporated into the spectroscopic accuracy, component compatibility, and working condition adaptability. The constraint loss weight is linked with the feature priority weight to ensure that high-priority constraints are met first, effectively improving the accuracy and reliability of the component matching model output results, so that the recommended component combination can meet both the matching requirements and the core constraints.

[0090] This invention provides an intelligent recommendation method for constructing a dichroic prism optical system. For each component recommendation combination, maximum and minimum corrections are applied based on the combination attributes to obtain a set of corrected experimental parameters, including: The recommended combination of components for a dichroic prism optical system is analyzed and classified by attributes to determine performance attributes, structural adaptability attributes, operating condition compatibility attributes, and characteristic parameters. Real-time priority weighting based on optical features Determine the priority order of various attributes and the dynamic boundary constraints; For each component, a recommended combination is made, and targeted maximum and minimum corrections are performed according to the attribute type. The correction parameters for each type of attribute are then summarized to form a set of correction experiment parameters.

[0091] In this embodiment, the deconstruction and classification of combined attributes refers to breaking down all attributes of the recommended combination of components and dividing them into three categories according to their function: performance attributes, structural adaptation attributes, and operating condition compatibility attributes. Each category of attributes contains several specific feature parameters to facilitate targeted correction. For example, after the attribute deconstruction of the recommended combination of components (filter F1 + prism P1), the performance attributes include the spectral band, transmittance, and spectral efficiency; the structural adaptation attributes include the installation dimensions, interface type, and stacking coaxiality; and the operating condition compatibility attributes include the temperature and humidity tolerance range, vibration tolerance threshold, and performance drift.

[0092] In this embodiment, the correction priority ranking refers to determining the correction priority of three types of attributes based on the real-time priority weights of optical features. Performance attributes have the highest correction priority, structural adaptation attributes are in the middle, and operating condition compatibility attributes have the lowest, ensuring that high-priority attributes are corrected accurately first. For example, in the real-time priority weights of optical features, the high priority weights corresponding to performance attributes... =0.65, the medium priority weight corresponding to the structural adaptation class attribute. =0.25, the low priority weight corresponding to the working condition compatibility class attribute. If the value is 0.1, then the priority order for correction is: performance-related attributes > structural adaptation-related attributes > working condition compatibility-related attributes.

[0093] In this embodiment, dynamic boundary constraints refer to the maximum and minimum correction boundaries set for the three types of attributes respectively. The boundary conditions are dynamically adjusted according to the priority weight of optical features to ensure that the correction parameters are within a reasonable range. For example, the core requirement performance upper limit for transmittance, a performance-type attribute, is 98%, the basic operating performance lower limit is 92%, and the higher priority weight... =0.65, for every 10% increase in weight, the boundary width narrows by 8%. The initial boundary width is 6% (98%-92%), so the current boundary width = 6%×(1-0.65×80%). The maximum correction boundary = 92%+2.88%, the minimum correction boundary = 92%. The maximum tolerance range for the installation dimensions of structural adaptation attributes is ±0.2mm, the minimum gap threshold is ±0.05mm, and the weight is medium priority. =0.25<0.5, correction range = 30%×(0.25 / 0.5)=15%, maximum correction boundary = ±0.05mm+15%×(0.2mm-0.05mm), minimum correction boundary = ±0.05mm; the extreme operating condition stability parameter range of the temperature tolerance range for the operating condition compatibility attribute is -10℃-50℃, the reference operating condition adaptation parameter is 0℃-40℃, and the low priority weight. =0.1. For every 10% increase in weight, the correction range widens by 5%. The initial correction range is 40℃ (40℃-0℃). Therefore, the current correction range = 40℃ × (1 + 0.1 × 50%), the maximum correction boundary = 0℃ to 42℃, and the minimum correction boundary = 0℃ to 40℃. Boundary width = initial boundary width × (1 - priority weight of the corresponding attribute × 0.8), maximum correction boundary = basic operating performance lower limit + boundary width, correction magnitude = 30% × (Pm / 0.5), where Pm is the medium priority weight corresponding to the structural adaptation class attribute (value range 0-0.5), and 30% is the baseline correction magnitude.

[0094] In this embodiment, directional maximization correction refers to parameter adjustments made to various attributes in a direction that improves the performance, compatibility, or adaptability of component combinations, so that the corrected parameters can maximize the satisfaction of core requirements. For example, the initial parameter of the performance attribute, spectral efficiency, is 90%, and after maximization correction in the direction of improving spectral efficiency, the corrected parameter is 95%; the initial parameter of the structural adaptation attribute, interface compatibility, is 85%, and after correction in the direction of maximizing assembly compatibility, the corrected parameter is 98%; the initial parameter of the operating condition compatibility attribute, humidity tolerance range, is 40%-60%, and after correction in the direction of expanding the adaptability range, the corrected parameter is 30%-70%.

[0095] In this embodiment, directional minimum correction refers to parameter correction for various attributes in a direction that meets the basic operational requirements of the component assembly, ensuring that the corrected parameters guarantee the minimum performance standard of the system. For example, the initial parameter for the performance attribute of spectral accuracy deviation is ±3nm, and after correction in a direction that meets the minimum optical performance, the corrected parameter is ±2nm (ensuring basic accuracy requirements); the initial parameter for the structural adaptation attribute of stacking coaxiality error is 0.1mm, and after correction in a direction that minimizes assembly error, the corrected parameter is 0.05mm; the initial parameter for the operating condition compatibility attribute of vibration tolerance threshold is 5g, and after correction in a direction that ensures stable operation under the baseline operating condition, the corrected parameter is 3g.

[0096] In this embodiment, the modified experimental parameter set pair refers to the parameter set pair consisting of the maximal modified experimental parameter set and the minimum modified experimental parameter set.

[0097] The beneficial effects of the above technical solution are as follows: by classifying and prioritizing the attributes of the recommended component combinations, setting dynamic boundary constraints, and performing directional maximum and minimum corrections, the resulting set of corrected experimental parameters can comprehensively cover the reasonable range of parameter variations. This not only considers the optimal performance of the component combinations but also ensures basic operational requirements, providing a rich and reliable parameter foundation for subsequent optimal value calculations and improving the comprehensiveness and accuracy of parameter optimization.

[0098] This invention provides an intelligent recommendation method for constructing a dichroic prism optical system, which calculates the optimal values ​​for the initial experimental parameter set and the corrected experimental parameter set for each component recommendation combination, including: Calculate the comprehensive base score for each component recommendation combination based on the corresponding initial experimental parameter set, maximum correction experimental parameter set, and minimum correction experimental parameter set; The basic comprehensive score for different sets of experimental parameters is corrected to obtain the corrected comprehensive score, and the consistency coefficient of the recommended combination of each component is calculated. If the consistency coefficient is greater than the preset coefficient, the corrected comprehensive score is weighted and calculated to obtain the optimal value. Otherwise, the score corresponding to the parameter set with the highest score after correction is retained first, and the items below the indicator threshold are fine-tuned by combining the individual scores of each evaluation indicator in the corresponding parameter set. After fine-tuning, the comprehensive score is recalculated as the optimal value.

[0099] In this embodiment, the performance fit score is... The score range is 0 to 1; Structural compatibility score The score ranges from 0 to 1.2, with a baseline assembly pass rate of 95%. Operating condition stability score The score range is 0 to 1; Basic Comprehensive Score ,in, The weights for performance adaptability, structural compatibility, and operating condition stability are respectively assigned, and =1, and thus we can obtain the combined scores of the initial experimental parameter set, the maximum correction experimental parameter set, and the minimum correction experimental parameter set, which are represented by R1, R2, and R3 respectively; Corrections to R1, R2, and R3 ,in, The corrected score; For the working condition adaptation factor, and =Preset experiment success rate, with a value ranging from 0.7 to 1.0. The value ranges from 1.0 to 1.2. The value ranges from 1.0 to 1.15.

[0100] In this embodiment, the consistency coefficient is calculated as follows: Furthermore, the preset coefficient is 0.8. Based on a large amount of experimental data, the weighted calculation of the comprehensive score is adjusted. The corresponding weight is 0.5. The corresponding weight is 0.3. The corresponding weight is 0.2.

[0101] In this embodiment, the indicator thresholds are set according to the requirements standards, such as performance adaptability threshold = 0.85, structural compatibility threshold = 0.9, and operating condition stability threshold = 0.85. The parameters corresponding to the individual scores below the thresholds are slightly adjusted, and the comprehensive score is recalculated to ensure that the optimal value can reflect the stable performance of the component combination.

[0102] The beneficial effects of the above technical solution are: by calculating the comprehensive basic score and the corrected comprehensive score, and combining the consistency coefficient to judge the stability of the score, the optimal value is determined by weighted calculation or secondary fine-tuning. This not only considers the comprehensive performance of different parameter sets, but also ensures the reliability and stability of the optimal value through consistency judgment and fine-tuning, providing a precise quantitative basis for the selection of the final component recommendation combination.

[0103] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligently recommending solutions for constructing a dichroic prism optical system, characterized in that, include: Step 1: Obtain the explicit and implicit granular features in the optical requirements, and perform correlation mining on the explicit and implicit granular features to construct an optical feature matrix; Step 2: Establish a dynamic attenuation function for optical feature priority that adapts to the core requirements of the dichroic prism. Use a hierarchical optical feature encoder to split the optical feature matrix into layers according to the feature priority dimension. Combine the pre-built optical component knowledge graph to map the split matrix features to the optical component semantic space. Step 3: Input the mapping results into the component matching model built by the heterogeneous graph neural network of optical components, and optimize the output matching results of the component matching model with the core optimization objectives of the dichroic prism system's spectral accuracy, component compatibility, and working condition adaptability to obtain the recommended component combination of the dichroic prism optical system. The recommended component combination includes optical processing components and optical construction components, and the number of internal components of the optical processing components and optical construction components in different recommended component combinations is not unique. The heterogeneous graph neural network of optical components incorporates the performance constraints of the optical adaptation learning framework. Step 4: Mark and conduct preliminary experiments for each recommended combination of components to obtain an initial experimental parameter set. Then, perform maximum and minimum corrections on each recommended combination of components according to the combination attributes to obtain a corrected experimental parameter set pair. Finally, calculate the initial experimental parameter set and the corrected experimental parameter set pair for each recommended combination of components to determine the optimal value. Step 5: Select the component recommendation combination corresponding to the final value from all the optimal values ​​as the final recommended combination, and build the dichroic prism optical system.

2. The intelligent recommendation method for constructing a dichroic prism optical system according to claim 1, characterized in that, The explicit and implicit granular features are correlated and mined to construct an optical feature matrix, including: The pre-trained optical requirement-feature association topology network is invoked to perform targeted mining of the potential mapping associations between the explicit and implicit granular features corresponding to the optical requirements of the dichroic prism, and to generate a mining set for each explicit granular feature. And the mining set for generating each latent granular feature. The explicit granularity features are concrete optical performance index features, and the implicit granularity features are potential demand dimension features. Mining set of any set of explicit granular features Mining set with latent granular features Analyze the content union coverage of the two sets of mining sets. and overlapping dimensions The strength of demand transmission between the explicit and implicit features of corresponding groups is quantified to obtain the corresponding demand influence weight values, where, ; Using the explicit granularity feature dimension, the implicit granularity feature dimension, and the demand influence weight dimension as three-dimensional coordinate axes, a three-dimensional optical feature matrix exclusive to the color separation prism is constructed. Each element in the three-dimensional optical feature matrix uniquely corresponds to a set of relationships between explicit and implicit granular features, and the element carries the demand influence weight value corresponding to the set of relationships.

3. The intelligent recommendation method for constructing a dichroic prism optical system according to claim 1, characterized in that, Establish a dynamic attenuation function for optical feature priority that adapts to the core requirements of dichroic prisms, including: Differentiated attenuation coefficients are set according to the three stages of the working cycle of the dichroic prism, and the timeliness factor of the staged demand is calculated. ,in, These represent the start times of each stage. These are the attenuation coefficients for each stage, determined by the time allocation ratio of the dichroic prism's working tasks. Based on dynamic adjustment of the demand deviation rate, the real-time demand urgency factor is obtained. ,and ,in, For the k-th demand dimension, the real-time performance parameter is denoted as . This represents the standard value for the k-th demand dimension; This is the urgency gain coefficient; Determine the anomaly adaptation redundancy factor ,in, This represents the current adaptation margin for the k-th demand dimension; This represents the maximum allowable adaptation margin for the k-th requirement dimension. It is a local minimum, and ; Based on factor interaction terms The coupling effect of the timeliness and urgency of the representation on the constructed dynamic decay function To obtain the intermediate priority weight And obtain the optical feature priority weights corresponding to the k-th demand dimension at time t. ,in, This is the coupling strength coefficient; This is a redundancy correction factor; This is a correction factor; Let be the initial priority weight for the k-th demand dimension; Determine the Euclidean distance between two adjacent weights The update cycle is adaptively adjusted to achieve real-time adaptation of priority weights to dynamic changes in requirements.

4. The intelligent recommendation method for constructing a dichroic prism optical system according to claim 1, characterized in that, By combining a pre-built optical component knowledge graph, the split matrix features are mapped to the optical component semantic space, including: A hierarchical optical feature encoder with adaptive encoding parameter configuration is constructed. The encoder contains three dedicated encoding sub-layers with high, medium, and low priorities, and the encoding parameters of each sub-layer are based on the priority weights of optical features. Dynamic configuration; The high, medium, and low priority feature subsets are input into their respective coding sub-layers, and after coding processing, hierarchical coded feature sets are obtained. Simultaneously, the information entropy of each encoded feature set is calculated. ,like If so, the corresponding encoding sublayer will be re-encoded; An optical component knowledge graph with multiple types of associated edges is constructed. The optical component knowledge graph includes four types of nodes: optical processing component nodes, optical construction component nodes, adaptation constraint nodes, and working condition associated nodes, and three types of weighted associated edges: component-component compatibility edges, component-constraint adaptation edges, and component-working condition associated edges. Fusion hierarchy priority matching factor Calculate the basic semantic similarity value ,in, This is the feature vector of the j2th component node in the optical component knowledge graph; Based on multi-constraint similarity pairs The final semantic similarity is obtained by calibration. The node feature vectors are extracted and weighted according to similarity weights to generate a set of semantic mapping vectors for optical components with hierarchical encoding features. This completes the mapping of features to the semantic space of optical components.

5. The intelligent recommendation method for constructing a dichroic prism optical system according to claim 1, characterized in that, The loss function of the heterogeneous graph neural network of the optical components ,in, Cross-entropy loss for component matching tasks; The squared loss for the three types of constraints, To constrain the loss weights, The constraint loss gain coefficient, and The high priority weights corresponding to the spectral accuracy. To ensure component compatibility with corresponding medium priority weights, Adapt the corresponding low-priority weights to the operating conditions; These are constraints on spectral accuracy, component compatibility, and operating condition adaptability.

6. The intelligent recommendation method for constructing a dichroic prism optical system according to claim 1, characterized in that, For each component, the recommended combination is adjusted for maximum and minimum values ​​based on the combination attributes to obtain a set of adjusted experimental parameters, including: The recommended combination of components for a dichroic prism optical system is analyzed and classified by attributes to determine performance attributes, structural adaptability attributes, operating condition compatibility attributes, and characteristic parameters. Real-time priority weighting based on optical features Determine the priority order of various attributes and the dynamic boundary constraints; For each component, a recommended combination is made, and targeted maximum and minimum corrections are performed according to the attribute type. The correction parameters for each type of attribute are then summarized to form a set of correction experiment parameters.

7. The intelligent recommendation method for constructing a dichroic prism optical system according to claim 1, characterized in that, For each component, the initial experimental parameter set and the corrected experimental parameter set are calculated to determine the optimal values, including: Calculate the comprehensive base score for each component recommendation combination based on the corresponding initial experimental parameter set, maximum correction experimental parameter set, and minimum correction experimental parameter set; The basic comprehensive score for different sets of experimental parameters is corrected to obtain the corrected comprehensive score, and the consistency coefficient of the recommended combination of each component is calculated. If the consistency coefficient is greater than the preset coefficient, the corrected comprehensive score is weighted and calculated to obtain the optimal value. Otherwise, the score corresponding to the parameter set with the highest score after correction is retained first, and the items below the indicator threshold are fine-tuned by combining the individual scores of each evaluation indicator in the corresponding parameter set. After fine-tuning, the comprehensive score is recalculated as the optimal value.

8. The intelligent recommendation method for constructing a dichroic prism optical system according to claim 1, characterized in that, The optical processing component is used to filter and shape the incident light to obtain the beam to be separated into colors, and the optical assembly component is used to perform beam separation processing on the beam to be separated into colors.

Images