A parameter optimization method for SPR grating type sensor based on FPCL-DCNN

Abstract

This invention discloses a parameter optimization method for SPR grating-type sensors based on FPCL-DCNN, relating to the field of sensors. Specifically, it involves treating an unknown gas or liquid as the analyte, calculating the elements that react with the surface of its bio-based membrane and the metal layer material of the sensor, and determining whether each element exceeds its respective set threshold. If at least one element exceeds the threshold, it indicates no reaction, and the metal layer material of the sensor is replaced, and the values ​​of the reacting elements are recalculated. Otherwise, initial parameters are set for both, and optimization is performed using a DCNN model: feature extraction of the initial parameters is performed through multi-channel convolution of the DCNN; a spatial-parameter correlation graph is constructed; an adaptive attention mechanism is used to assign weights to the edges in the correlation graph, and the weights are adaptively adjusted; finally, the optimal values ​​of each parameter are output. The parameter prediction results are adjusted by calculating the deviation from the actual optimal values. This invention ensures that the sensor maintains high accuracy under changing environmental conditions.

Description

Technical Field

[0001] This invention relates to the fields of sensors and artificial intelligence, specifically a parameter optimization method for SPR grating type sensors based on FPCL-DCNN (Functionalized Physical Chemical Layer - Deep Convolutional Neural Network). Background Technology

[0002] Kretschmann-type SPR (Surface Plasmon Resonance) grating sensors are widely used in the field of optical sensing. Due to their high sensitivity, they are used for refractive index detection and molecular interaction research, and are widely applied in environmental monitoring, biomedical detection and chemical analysis.

[0003] SPR grating sensors utilize surface plasmon resonance excited by a metal film to sensitively respond to minute changes in the ambient refractive index. Current research focuses on improving material selection, structural optimization, and optical coupling techniques, such as choosing gold or silver films as materials, or enhancing signal sensitivity by adjusting film thickness and incident angle. However, these designs still face several limitations:

[0004] First, it is difficult to balance sensitivity and stability when choosing materials for metal films. Silver films have high sensitivity but are easily oxidized, while gold films have good stability but low sensitivity.

[0005] Secondly, the Kretschmann structure is highly dependent on the film thickness and incident angle, which limits its response range and restricts its flexibility in dynamic detection.

[0006] In addition, SPR signals are susceptible to interference from changes in ambient light and temperature, which can affect detection accuracy. While multilayer membrane structures can enhance detection performance, they may also lead to decreased signal repeatability due to instability at the interlayer interfaces.

[0007] Therefore, although SPR sensors have important application value, there is still room for improvement in terms of material innovation, structural design, and suppression of environmental interference.

[0008] SPR sensors are widely used in critical scenarios such as gas detection, toxicity analysis, and pathogen identification. Furthermore, with the increasing demands of modern industry and public safety, these sensors face higher requirements in terms of sensitivity, real-time response, and adaptability, particularly in high-precision detection, dynamic change monitoring, and long-term stability in complex environments. To meet these demands, the development of SPR sensors must not only improve the recognition capability of single signals but also achieve simultaneous detection and analysis of multiple signals to satisfy the needs of multifunctional applications.

[0009] Driven by artificial intelligence and large-scale modeling technologies, the detection capabilities and application potential of SPR grating sensors have been significantly enhanced. AI algorithms not only rapidly process SPR sensor signal data and extract key features, thereby improving the signal-to-noise ratio and response speed, but also enable the sensor to more accurately distinguish target signals in complex backgrounds. The introduction of large-scale models has also greatly expanded the application scope of detection, moving beyond single-analysis modes to construct multi-dimensional detection models through multi-level, cross-modal data learning, thus achieving more accurate and robust signal recognition.

[0010] Furthermore, deep learning technology endows SPR sensors with adaptive analysis capabilities, enabling them to dynamically adjust sensor parameters based on different detection environments, ensuring they are always in optimal working condition. During the sensor design process, computer models are used to simulate and analyze the optical and physicochemical properties of the target material, thereby accurately determining the most suitable sensor structural parameters. This adaptive mechanism not only improves the long-term stability of the sensor in complex environments but also enhances its sensitivity and real-time response capabilities, further expanding the application potential of SPR sensors in environmental monitoring, biomedicine, chemical analysis, and other fields. Summary of the Invention

[0011] To address the problem of the inability to precisely control the interaction between the analyte and the sensor surface in existing technologies, and the lack of core components for detecting surface plasmon resonance (SPR) of unknown gases or liquids, this invention proposes a parameter optimization method for an SPR grating sensor based on FPCL-DCNN. This method enables the design of parameters such as "biofilm thickness, metal film material and thickness, number of teeth, spacing, incident angle, and wavelength" of the core SPR component, as well as a method for evolving detection performance based on these parameters.

[0012] The specific steps of the SPR grating sensor parameter optimization method are as follows:

[0013] Step 1: An unknown gas or liquid is used as the analyte molecule. Calculate the elemental values ​​of the reaction between the surface of its bio-based membrane and the metal layer material of the SPR sensor;

[0014] The elements include: active groups on the surface of the bio-based membrane of the analyte molecule X. After chemical bonding with the metal layer material, the modification concentration of the analyte molecule and the remaining concentration of active groups B on the base film surface. ;

[0015] Analytical molecules in gaseous or liquid form The number of sites n on the base film surface after physical adsorption and the degree of reaction on the base film surface, i.e., surface coverage. ;

[0016] After chemical and physical modification, the test molecule Residual concentration in the medium channel ; and the comprehensive applicability parameter S of the biofilm.

[0017] Modification concentration The calculation is as follows:

[0018]

[0019] The molecule to be tested The initial concentration, The equilibrium constant is the reaction equilibrium constant. Indicates time.

[0020] Remaining concentration for:

[0021]

[0022] In the formula, Active groups on the surface of bio-based membranes The initial concentration.

[0023] The number of sites n is calculated as follows:

[0024]

[0025]

[0026] N is the molecule to be tested. The total number of physical adsorption sites on the base film surface. gaseous molecule to be tested The partial pressure, where C is the liquid molecule to be tested. The equilibrium concentration. K is the adsorption equilibrium constant: ; The molecule to be tested The concentration of products generated by the reaction of active group B on the base film surface with the metal layer material;

[0027] Surface coverage :

[0028]

[0029]

[0030] Remaining concentration :

[0031]

[0032] In the formula, Let Avogadro's constant be 1. It is the volume of the entire medium channel of the sensor.

[0033] Comprehensive applicability parameter S of bio-based membrane:

[0034]

[0035] In the formula, , , These correspond to the functional group reaction matching degree F and the physicochemical reaction matching degree, respectively. The weighting parameter for the similarity and compatibility matching degree L. Among them, the functional group reaction matching degree... :

[0036]

[0037] In the formula, It is the equilibrium constant of functional group reactions. To estimate the molecule to be tested The concentration of functional groups can be reflected in the middle. This represents the concentration of functional groups on the surface of the biofilm. It is a normalization constant, which makes The value is between 0 and 1. The number of specific functional groups contained on the surface of the biofilm; The number of functional group ligands in the analyte X that can bind to the properties of the base membrane.

[0038] Physicochemical reaction matching degree :

[0039]

[0040] In the formula, It is the slope of the curve; The measured physical adsorption free energy; This is the median adsorption free energy. When the adsorption of the analyte between the biofilm and the membrane is observed to be at a moderate level, .

[0041] Similarity and compatibility matching degree L:

[0042]

[0043] In the formula, Indicates the polarity of the base film. Representative of the substance to be tested polarity, This indicates the difference in polarity between the two. It is the predefined maximum polarity difference.

[0044] Step 2: Determine whether the value of each element exceeds its respective set threshold. If so, proceed to Step 3; otherwise, it indicates that the molecule to be tested... The surface of the bio-based membrane did not react with the metal layer material of the SPR sensor. After replacing the metal layer material of the sensor, the values ​​of the reacting elements were recalculated.

[0045] Step 3: Set the molecule to be tested The initial parameters of the metal layer material of the SPR sensor were determined and optimized using the DCNN optimization model to improve the response performance.

[0046] Initialization parameters include initial biofilm thickness The reflectivity R of the reactive metal film material and thickness, and the number of teeth in the grating parameters. and the gap between the concave and convex surfaces and the wavelength of the incident light source With the angle of incidence :

[0047] Bio-based membrane thickness definition:

[0048]

[0049] in, It is the surface area of ​​the biofilm. These represent the mass and density of the bio-based membrane, respectively.

[0050] Reflectivity R is expressed as:

[0051]

[0052] The transmission matrix M is:

[0053]

[0054] The elements of the matrix represent the optical properties of each layer;

[0055] Number of teeth It is the number of grooves in a grating within a certain length, and the spacing between the grooves and the convexities. There is an inverse proportional relationship, expressed as:

[0056]

[0057] Concave-convex spacing :

[0058]

[0059] For diffraction order, The wavelength of the light source, These are the incident angle and diffraction angle corresponding to this light source, respectively.

[0060] Step 4: Design and optimize the multi-channel convolutional noise reduction feature extraction layer of the DCNN model to extract multi-channel features from the initial parameters;

[0061] Input features: biofilm thickness, metal film material, metal film thickness, number of grating teeth, concave-convex spacing, incident angle, and incident light wavelength. Convolution operations are applied, gradually increasing the feature dimension through multiple layers of 1D convolutions, and combined with a noise reduction layer. The output encoded features are as follows:

[0062]

[0063] In the formula, It is the first The convolutional output of the layer, It is the first The convolutional kernel weights of the layer, Represents a local region of the input data. It is the first Layer bias terms, It is an activation function. It represents the number of convolution kernels.

[0064] Step 5: Construct a space-parameter relationship graph using the initial parameters;

[0065] Each input parameter (biofilm thickness, metal film material, metal film thickness, number of grating teeth, concave-convex spacing, incident angle, incident light wavelength) is defined as a node in the graph, and the edge weight between different nodes represents the correlation strength between the parameters.

[0066] The calculation formula for DCNN is as follows:

[0067]

[0068] In the formula, It involves adding a self-connected adjacency matrix. It is an adjacency matrix The degree matrix, Indicates the first Feature representation of layer nodes, It is the first Layer-learnable weight matrix.

[0069] Step 6: Use an adaptive attention mechanism to assign weights to the edges between nodes with different parameters in the association graph, and perform adaptive weight adjustment;

[0070] The adaptive attention mechanism includes a local attention mechanism and a global attention mechanism;

[0071] First, the local attention mechanism pools the multi-channel feature map output by the extraction layer to obtain global statistical information for each channel.

[0072] Then, a fully connected layer is used to compress global statistics into channel-weighted coefficients, and the contribution of different channels is recalibrated through attention weights.

[0073] Channel weighting coefficient The formula is as follows:

[0074]

[0075] In the formula, W is the weight matrix of the fully connected layer, and pool is the global pooling operation. It represents the total number of all channels.

[0076] Finally, the global attention mechanism utilizes pooling and channel weighting coefficients to fuse features from multi-layer DCNN and convolutional layers, obtaining a multi-dimensional feature representation while suppressing noise or unimportant information. The formula is as follows:

[0077]

[0078] In the formula, L is the number of pooling layers.

[0079] Step 7: Optimize the model. The output layer of the DCNN is designed as a linear activation layer, outputting the optimal values ​​of each parameter. .

[0080] The formula for the linear combination of output is:

[0081]

[0082] In the formula, and These are the weight matrix and bias of the output layer, respectively, and Z is the feature representation after multi-layer fusion.

[0083] Step 8: Use the loss function that minimizes the mean squared error (MSE), the mean absolute error (MAE), and... The coefficient of determination measures the optimal value. The deviation from the actual optimal value is used to adjust the parameters and predict the results.

[0084] The formula for calculating the mean squared error (MSE) is as follows:

[0085]

[0086] The molecule to be tested The initial parameters of the bio-based membrane, the initial parameters of the metal layer material of the SPR sensor, and the actual values ​​of each element involved in the reaction between the two. The molecule to be tested Initial parameters of the bio-based membrane, initial parameters of the metal layer material of the SPR sensor, and predicted values ​​of each element involved in the reaction between the two. It refers to the number of samples.

[0087] The formula for calculating the Mean Absolute Error (MAE) is:

[0088]

[0089] The formula for calculating the coefficient of determination is:

[0090]

[0091] The molecule to be tested The initial parameters of the bio-based membrane, the initial parameters of the metal layer material of the SPR sensor, and the average values ​​of each element in the reaction.

[0092] The advantages of this invention are:

[0093] (1) The present invention provides a parameter optimization method for SPR grating type sensor based on FPCL-DCNN. By performing physical and chemical dual modification on the substance to be tested, a specific bio-based membrane is designed and functional groups and physical and chemical modifications are introduced to enhance the binding force between the molecule to be tested and the sensor surface.

[0094] (2) This invention provides a parameter optimization method for an SPR grating-type sensor based on FPCL-DCNN. According to the reaction rate and molecular size of different analytes, the optimal biomembrane is designed for each analyte, enabling the sensor to achieve rapid response and stable detection results in a shorter time. Through precise control of the membrane layer, this invention ensures that the sensor maintains high accuracy under changing environmental conditions.

[0095] (3) The present invention provides a parameter optimization method for SPR grating type sensor based on FPCL-DCNN, which can automatically optimize the structural design of the sensor in real time according to the changes of the substance to be measured and the design requirements of the sensor, thereby improving the detection sensitivity and accuracy of the SPR sensor for specific substances. Attached Figure Description

[0096] Figure 1 This is a flowchart of a parameter optimization method for an SPR grating sensor based on FPCL-DCNN according to the present invention;

[0097] Figure 2 This is a schematic diagram of the structure of the DCNN multimodal convolutional neural network of the present invention;

[0098] Figure 3This is a flowchart of the DCNN multimodal convolutional neural network algorithm of the present invention;

[0099] Figure 4 This is a schematic diagram of the results of five selected sets of experimental data in an embodiment of the present invention. Detailed Implementation

[0100] The specific implementation method of the present invention will be further described in detail below with reference to the accompanying drawings.

[0101] This invention provides a parameter optimization method for SPR grating-type sensors based on FPCL-DCNN (Functionalized Physical Chemical Layer-Deep Convolutional Neural Network). It innovatively applies DCNN to the design and optimization of SPR sensors, using the relevant physicochemical properties of the analyte and the sensor's structural parameters as input features. This input is passed to the DCNN model to further learn the dependencies between various sensor parameters. Each parameter is treated as a node in a graph, and the edges between nodes represent their interactions. This model can automatically identify and fuse the relationships between different parameters to find the optimal parameter combination. Simultaneously, an adaptive attention mechanism is introduced, using a global pooling layer to weight the features output from convolution and graph convolution, automatically adjusting the influence of different features. Finally, the DCNN outputs the optimal SPR sensor parameters through a linear activation function in the output layer.

[0102] like Figure 1 As shown, the specific steps are as follows:

[0103] Step 1: An unknown gas or liquid is used as the analyte molecule. Calculate the elemental values ​​of the reaction between the surface of its bio-based membrane and the metal layer material of the SPR sensor;

[0104] The elements include: active groups on the surface of the bio-based membrane of the analyte molecule X. After chemical bonding with the metal layer material, the modification concentration of the analyte molecule and the remaining concentration of active groups B on the base film surface. ;

[0105] Analytical molecules in gaseous or liquid form The number of sites n on the base film surface after physical adsorption and the degree of reaction on the base film surface, i.e., surface coverage. ;

[0106] After chemical and physical modification, the test molecule Residual concentration in the medium channel ; and the comprehensive applicability parameter S of the biofilm.

[0107] Modification concentration The calculation is as follows:

[0108] First, calculate the molecule to be tested. With active groups on the base film surface degree of integration :

[0109]

[0110] The molecule to be tested The initial concentration, in units of , This is the reaction equilibrium constant, in units of . , Indicates time, in units of .

[0111] Then, based on the degree of binding Calculate the number of test molecules that bind to active group B on the base film surface. Modification concentration :

[0112]

[0113] Thus, the remaining concentration of active groups B on the base film surface was obtained. for:

[0114]

[0115] In the formula, Active groups on the surface of bio-based membranes The initial concentration, in units of .

[0116] Based on the molecule to be tested Modification concentration Calculate the molecule to be tested separately Physical adsorption coverage on the surface of gas or liquid base film ;

[0117] First, calculate the adsorption equilibrium constant K: ;

[0118] The molecule to be tested Concentration of products generated by the reaction of active group B on the base film surface with the metal layer material;

[0119] Then, the molecule to be tested is calculated. The number of sites n on the surface of the base film in a gas or liquid after physical adsorption;

[0120]

[0121]

[0122] N is the molecule to be tested. The total number of physical adsorption sites on the base film surface. gaseous molecule to be tested The partial pressure, where C is the liquid molecule to be tested. The equilibrium concentration.

[0123] Finally, the adsorption isotherms of the analyte molecules were calculated respectively. The degree of reaction on the surface of a gas or liquid substrate film, i.e., surface coverage. :

[0124]

[0125]

[0126] After chemical and physical two-layer modification, the analyte molecule was calculated. Residual concentration in the medium channel :

[0127]

[0128] In the formula, Let Avogadro's constant be 1. It is the volume of the entire medium channel of the sensor.

[0129] Comprehensive applicability parameter S of bio-based membrane:

[0130]

[0131] In the formula, , , These correspond to the functional group reaction matching degree F and the physicochemical reaction matching degree, respectively. The weighting parameter of the similarity and compatibility matching degree L.

[0132] Among them, with the molecule to be tested Functional group reaction matching degree in the biofilm of the reaction :

[0133]

[0134] In the formula, It is the equilibrium constant of functional group reactions. To estimate the molecule to be tested Concentration of reactive functional groups: ; The number of functional group ligands in the analyte X that can bind to the properties of the base membrane; The concentration of functional groups on the surface of bio-based membranes: ; The number of specific functional groups contained on the surface of the biofilm; The surface area of ​​the base film; It is a normalization constant, which makes The value is between 0 and 1.

[0135] The physicochemical reaction matching degree is calculated based on the change in the adsorption amount of analyte X on the surface of the biofilm at different temperatures with varying partial pressure or concentration. :

[0136]

[0137] In the formula, It is the slope of the curve; The measured physical adsorption free energy; This is the median adsorption free energy. When the adsorption of the analyte between the biofilm and the membrane is observed to be at a moderate level, .

[0138] Similarity and compatibility matching degree L:

[0139]

[0140] In the formula, Indicates the polarity of the base film. Representative of the substance to be tested polarity, This indicates the difference in polarity between the two. It is the predefined maximum polarity difference.

[0141] Step 2: Determine whether the value of each element exceeds its respective set threshold. If so, proceed to Step 3; otherwise, it indicates that the molecule to be tested... The surface of the bio-based membrane did not react with the metal layer material of the SPR sensor. After replacing the metal layer material of the sensor, the values ​​of the reacting elements were recalculated.

[0142] Step 3: Set the molecule to be tested The initial parameters of the metal layer material of the SPR sensor were determined and optimized using the DCNN optimization model to improve the response performance.

[0143] Initialization parameters include initial biofilm thickness The reflectivity R of the reactive metal film material and thickness, and the number of teeth in the grating parameters. and the gap between the concave and convex surfaces and the wavelength of the incident light source With the angle of incidence :

[0144] Bio-based membrane thickness definition:

[0145]

[0146] in, It is the surface area of ​​the biofilm. These represent the mass and density of the bio-based membrane, respectively.

[0147] Reflectivity R is expressed as:

[0148]

[0149] The transmission matrix M is:

[0150]

[0151] The elements of the matrix represent the optical properties of each layer;

[0152] Number of teeth It is the number of grooves in a grating within a certain length, and the spacing between the grooves and the convexities. There is an inverse proportional relationship, expressed as:

[0153]

[0154] Concave-convex spacing :

[0155]

[0156] For diffraction order, The wavelength of the light source, These are the incident angle and diffraction angle corresponding to this light source, respectively.

[0157] Step 4: Design and optimize the multi-channel convolutional noise reduction feature extraction layer of the DCNN model to extract multi-channel features from the initial parameters;

[0158] The design of Convolutional Denoising Autoencoders (CDAEs) involves applying convolutional operations to the input features (biofilm thickness, metal film material, metal film thickness, grating tooth count, concave-convex spacing, incident angle, and incident light wavelength) to map the features to multiple channels, thereby extracting features at different levels. Through self-supervised denoising learning, the encoder can reduce the impact of environmental noise or data errors, making the encoded features more robust during reconstruction.

[0159] Multi-channel convolution is performed, treating each input parameter as an independent channel. The feature dimension is gradually increased through multiple layers of 1D convolution, combined with a denoising layer, making the encoded features of the convolution output more expressive in high-dimensional space. This facilitates subsequent attention mechanisms in capturing the importance of different parameters. The specific convolution calculation formula is as follows:

[0160]

[0161] In the formula, It is the first The convolutional output of the layer, It is the first The convolutional kernel weights of the layer, Represents a local region of the input data. It is the first Layer bias terms, It is an activation function. It represents the number of convolution kernels.

[0162] Step 5: Construct a space-parameter relationship graph using the initial parameters;

[0163] Each input parameter (biofilm thickness, metal film material, metal film thickness, number of grating teeth, concave-convex spacing, incident angle, incident light wavelength) is defined as a node in the graph, and the edge weight between different nodes represents the correlation strength between the parameters.

[0164] Graph convolutional networks (DCNNs) perform computation by assigning weights to the relationships between nodes, thus modeling the correlation between sensor parameters. This allows the model to learn the importance of each node in the overall features. The calculation formula for DCNNs is as follows:

[0165]

[0166] In the formula, It involves adding a self-connected adjacency matrix. It is an adjacency matrix The degree matrix, Indicates the first Feature representation of layer nodes, It is the first Layer-learnable weight matrix.

[0167] Step 6: Use an adaptive attention mechanism to assign weights to the edges between nodes with different parameters in the association graph, and perform adaptive weight adjustment;

[0168] The adaptive attention mechanism includes a local attention mechanism and a global attention mechanism;

[0169] First, the local attention mechanism pools the multi-channel feature map output by the extraction layer to obtain global statistical information for each channel.

[0170] Then, a fully connected layer is used to compress global statistics into channel-weighted coefficients, and the contribution of different channels is recalibrated using attention weights. This step prioritizes the output features of the convolutional layer before parameter fusion, helping the model to focus more effectively on important parameters. The attention mechanism generates channel-weighted coefficients through global pooling and fully connected layers. The formula is as follows:

[0171]

[0172] In the formula, It is the weight matrix of the fully connected layer, and `pool` is the global pooling operation. It represents the total number of all channels.

[0173] Finally, a global attention mechanism is constructed. Typically, the feature information output by different convolutional layers has different levels, with deeper layers exhibiting more abstract feature representations. During multi-layer fusion, a global attention mechanism is introduced. This mechanism utilizes pooling and channel weighting coefficients to fuse the features from multiple DCNN layers and convolutional layers, obtaining a multi-dimensional feature representation while suppressing noise or unimportant information. This ensures that the model's contribution to each feature layer can be adaptively optimized before the final output, as shown in the following formula:

[0174]

[0175] In the formula, L is the number of pooling layers.

[0176] Step 7: Optimize the model. The output layer of the DCNN is designed as a linear activation layer, outputting the optimal values ​​of each parameter. .

[0177] To obtain numerical results for the optimal design parameters, the output layer of the model can be designed as a linear activation layer, outputting the optimal values ​​of each parameter. The formula for the linear combination output is:

[0178]

[0179] In the formula, and These are the weight matrix and bias of the output layer, respectively, and Z is the feature representation after multi-layer fusion.

[0180] Step 8: Use the loss function that minimizes the mean squared error (MSE), the mean absolute error (MAE), and... The coefficient of determination measures the optimal value. The deviation from the actual optimal value is used to adjust the parameters and predict the results.

[0181] The formula for calculating the mean-square error (MSE) is as follows:

[0182]

[0183] The molecule to be tested The initial parameters of the bio-based membrane, the initial parameters of the metal layer material of the SPR sensor, and the elements involved in the reaction between the two. The true value The molecule to be tested The initial parameters of the bio-based membrane, the initial parameters of the metal layer material of the SPR sensor, and the elements involved in the reaction between the two. Predicted value It refers to the number of samples.

[0184] The formula for calculating the Mean Absolute Error (MAE) is:

[0185]

[0186] The formula for calculating the coefficient of determination is:

[0187]

[0188] The molecule to be tested The initial parameters of the bio-based membrane, the initial parameters of the metal layer material of the SPR sensor, and the elements involved in the reaction between the two. ]average value.

[0189] Example:

[0190] This implementation case uses a dataset of 800 samples, including 160 samples each from biomedical molecules (liquid), agricultural chemical chiral molecules (liquid), environmental monitoring pollutant molecules (gas), food safety preservative molecules (liquid), and pharmacological toxicity molecules (gas). A random sampling method was used to select 120 samples from each of these five datasets as the training set, and the remaining 40 sets were used as the test set. Ultimately, a total of 600 samples were used for training, and a total of 200 samples were used for testing.

[0191] The specific process is as follows:

[0192] (1) The molecule to be tested Active groups on the surface of bio-based membranes Modifications related to chemical reactions:

[0193] Due to the molecule to be tested The biochemical and optical properties of the sample are unknown, and the plasmon resonance response of the SPR sensor is related to the refractive index of the medium. Furthermore, to accelerate the change in the refractive index of the medium and amplify the SPR signal, the analyte molecule is modified with an FPCL (Functionalized Physical Chemical Layer).

[0194] First, the molecules to be tested Chemical modification of the active group B on the surface of the biofilm was carried out. Since the parameters can be obtained experimentally or under known conditions, any set of data for each category is selected to illustrate the entire process. The specific experimental values ​​are shown below:

[0195] Biomedical Molecular Mechanics: , , ,when At that time, degree of binding .

[0196] Chiral molecular group of agricultural chemistry: , , ,when At that time, degree of binding .

[0197] Environmental monitoring pollutant molecular composition: , , ,when At that time, degree of binding .

[0198] Food safety preservative molecular group: , , ,when At that time, degree of binding .

[0199] Pharmacological toxicology molecular group: , , ,when At that time, degree of binding .

[0200] Substituting the above parameters yields the molecule to be tested. Modification concentration :

[0201] Biomedical Molecular Mechanics:

[0202] Chiral molecular group of agricultural chemistry:

[0203] Environmental monitoring pollutant molecular composition:

[0204] Food safety preservative molecular group:

[0205] Pharmacological toxicology molecular group:

[0206] Substituting the parameters yields the remaining concentration. for:

[0207] Biomedical Molecular Mechanics:

[0208] Chiral molecular group of agricultural chemistry:

[0209] Environmental monitoring pollutant molecular composition:

[0210] Food safety preservative molecular group:

[0211] Pharmacological toxicology molecular group:

[0212] (2) Introduce the Langmuir physical adsorption model:

[0213] To enhance adaptability to different types of molecules, improve the stability and density of molecular immobilization, and adjust the response time and intensity of the detection signal, the synergistic effect of physical adsorption and chemical modification is utilized. Based on the differences between target molecules and interfering molecules, the detection selectivity is improved. The Langmuir physical adsorption model is then introduced.

[0214] Based on experimental conditions, the molecule to be tested... The total number of physical adsorption sites on the base film surface is , has been molecule The number of adsorption sites occupied is The adsorption equilibrium constant is The equilibrium partial pressure of gas molecules is uniformly denoted as The equilibrium concentration of liquid molecules is uniformly recorded as 0.4 mol / L.

[0215] Calculate the surface coverage based on the Langmuir adsorption isotherm.

[0216]

[0217]

[0218] Adsorbed molecules The quantity (expressed as amount of substance or number of molecules) n is:

[0219]

[0220] Substituting into the formula, we can obtain the amount of each group of molecules adsorbed:

[0221] Chemical reaction analysis reveals that... Substituting these values ​​into the formula, we get:

[0222] The biomedical molecular genome is:

[0223] Chiral molecular group of agricultural chemistry:

[0224] Environmental monitoring pollutant molecular composition:

[0225] Food safety preservative molecular group:

[0226] Pharmacological toxicology molecular group:

[0227] After chemical and physical modification, the test molecule Residual concentration in the medium channel The dynamic expression that changes over time is:

[0228] The biomedical molecular genome is:

[0229]

[0230] Chiral molecular group of agricultural chemistry:

[0231]

[0232] Environmental monitoring pollutant molecular composition:

[0233]

[0234] Food safety preservative molecular group:

[0235]

[0236] Pharmacological toxicology molecular group:

[0237]

[0238] (3) Identify the molecule to be tested Component Y of the reacting bio-based membrane:

[0239] First, calculate the functional group reaction matching degree F:

[0240] Determining the concentration of functional groups on the surface of bio-based membranes :

[0241] The density of specific functional groups on the surface of biofilms was determined using X-ray photoelectron spectroscopy (XPS), a chemical analytical method. XPS analysis revealed the presence of specific functional groups on the surface of the biofilm. A specific functional group, basement membrane surface area ,but .

[0242] Estimate the concentration of reactive functional groups in the analyte :

[0243] For small molecule compounds, their molecular structure can be determined using nuclear magnetic resonance (NMR), thus clarifying the types and numbers of reactive functional groups. For macromolecules, the average concentration of reactive functional groups can be estimated by combining methods such as amino acid sequence analysis or glycosylation analysis.

[0244] For different analytes X, the concentration of functional groups can be reflected. for:

[0245] Biomedical Molecular Mechanics: The biomedical molecules to be tested contain A reactive functional group, then .

[0246] Chiral molecular group of agricultural chemicals: The chiral molecules of agricultural chemicals to be tested contain A reactive functional group, then .

[0247] Environmental monitoring pollutant molecular composition: The environmental monitoring pollutant molecules to be tested contain A reactive functional group, then .

[0248] Food safety preservative molecular group: The food safety preservative molecules to be tested contain 3 A reactive functional group, then .

[0249] Pharmacological toxicity molecular group: The pharmacological toxicity molecules to be tested contain 7 A reactive functional group, then .

[0250] Calculate the functional group reaction matching degree It characterizes the binding rate of the reactive functional groups of the analyte molecules in the base membrane, and the unit problem is solved by unifying the concentration.

[0251] Known functional group reaction equilibrium constant Normalization constant Then, according to the formula, we can calculate:

[0252] Chiral molecular group of agricultural chemistry:

[0253]

[0254] Chiral molecular group of agricultural chemistry:

[0255]

[0256] Environmental monitoring pollutant molecular composition:

[0257]

[0258] Food safety preservative molecular group:

[0259]

[0260] Pharmacological toxicology molecular group:

[0261]

[0262] Next, the degree of physicochemical reaction matching is calculated. :

[0263] First, measure the physical adsorption free energy. :

[0264] The adsorption amount of the analyte on the surface of the biofilm was measured at different temperatures as a function of partial pressure or concentration. Langmuir adsorption isotherms were determined. The heat of adsorption was calculated at different temperatures using a variation of the Clausius-Clapeyron equation, and the adsorption free energy was calculated using thermodynamic formulas to determine the physical adsorption free energy. as follows:

[0265] Chiral molecular group of agricultural chemistry:

[0266] Chiral molecular group of agricultural chemistry:

[0267] Environmental monitoring pollutant molecular composition: ,

[0268] Food safety preservative molecular group: ,

[0269] Pharmacological toxicology molecular group: ,

[0270] Known According to the formula, we can calculate:

[0271] Chiral molecular group of agricultural chemistry:

[0272]

[0273] Chiral molecular group of agricultural chemistry:

[0274]

[0275] Environmental monitoring pollutant molecular composition:

[0276]

[0277] Food safety preservative molecular group:

[0278]

[0279] Pharmacological toxicology molecular group:

[0280]

[0281] Next, calculate the similarity and compatibility matching degree L:

[0282] Known The polarity of bio-based membranes According to the calculation based on the public announcement, we can obtain:

[0283] Chiral molecular group of agricultural chemistry:

[0284]

[0285] Chiral molecular group of agricultural chemistry:

[0286]

[0287] Environmental monitoring pollutant molecular composition:

[0288]

[0289] Food safety preservative molecular group:

[0290]

[0291] Pharmacological toxicology molecular group:

[0292]

[0293] Finally, the comprehensive applicability parameter S of the bio-based membrane was calculated and determined:

[0294]

[0295] In the formula, , , These correspond to the functional group reaction matching degree F and the physicochemical reaction matching degree, respectively. The weight parameters of the similarity and compatibility matching degree L can be determined later through a deep neural network model.

[0296] Known =0.3, which can be calculated using the formula:

[0297] Chiral molecular group of agricultural chemistry:

[0298]

[0299] Chiral molecular group of agricultural chemistry:

[0300]

[0301] Environmental monitoring pollutant molecular composition:

[0302]

[0303] Food safety preservative molecular group:

[0304]

[0305] Pharmacological toxicology molecular group:

[0306]

[0307] (4) Define the initial model parameters:

[0308] First, initialize the biofilm thickness. Biofilm thickness Indirectly affecting the analyte molecules by changing the interfacial refractive index. Changes in the surface plasma resonance wave of the SPR sensor;

[0309] Next, the material and thickness of the metal film are initialized. ; number of grating teeth Concave-convex spacing =200nm; incident light source wavelength With the angle of incidence .

[0310] SPR sensors typically use gold or silver as the metal layer material. The refractive index and dielectric constant of different materials will affect the sensitivity of the sensor. Common metal refractive index and absorption coefficient data are shown in Table 1:

[0311] Table 1

[0312] Metal Refractive index (n) Absorption coefficient (k) Wavelength range (nm) Gold n≈0.47-1.92 k≈1.47-6.2 400-800 Silver n≈0.04-0.18 k≈1.5-5.5 400-800 Copper n≈0.27-1.43 k≈2.64-5.0 400-800 Aluminum n≈0.65-1.73 k≈4.5-9.1 400-800 Platinum n≈1.75-2.32 k≈4.2-6.0 400-800 Nickel n≈1.5-2.4 k≈3.6-5.8 400-800 Palladium n≈1.2-2.5 k≈4.0-6.0 400-800 Chromium n≈2.4-3.5 k≈3.9-5.8 400-800 Titanium n≈2.0-2.8 k≈2.8-3.9 400-800 Iron n≈2.5-3.3 k≈3.0-4.8 400-800

[0313] The thickness of the metal film affects the sensitivity and peak position of the SPR signal, and the optimal metal film thickness can be analyzed using the reflectivity formula:

[0314] Reflectivity was calculated using the Abélès matrix method in the test material-biofilm-metal film structure. This method represents the optical properties of each layer using a matrix, and then multiplies them layer by layer to obtain the reflectivity of the entire structure.

[0315] The characteristic matrix of each layer j can be written as:

[0316]

[0317] In the formula, It is the phase delay of light in the j-th layer. Let be the refractive index of the j-th layer. For thickness, For angle, For S-polarized light, or For p-polarized light, It represents the imaginary unit.

[0318] The overall transmission matrix M can be expressed as:

[0319]

[0320] Based on this, the reflectivity R can be expressed as:

[0321]

[0322] Then, initialize and define the grating parameters (number of teeth). and the gap between the concave and convex surfaces )

[0323] The structural parameters of a grating affect the coupling efficiency of light, which can be described by the grating diffraction equation:

[0324]

[0325] Adjust the number of grating teeth and the gap between the concave and convex surfaces To achieve optimal coupling angle and sensitivity. Number of teeth. It is the number of grooves in a grating within a certain length, and the spacing between the grooves and the convexities. There is an inverse relationship; at a certain specific wavelength Reaching a specific launch angle The spacing between the protrusions and concave sections can be adjusted. To satisfy the grating equation:

[0326]

[0327] Adjust the number of grating teeth and the gap between the concave and convex surfaces This allows for the achievement of optimal coupling angle and sensitivity.

[0328] Finally, the incident light source wavelength is initialized and defined. With the angle of incidence :

[0329] Incident light source wavelength With the angle of incidence The position of the SPR resonance peak has a significant impact, being the critical angle for total internal reflection when light travels from a high-refractive-index medium to a low-refractive-index medium. Defined as:

[0330]

[0331] in, and Let be the refractive index of the incident medium and the metal layer, respectively.

[0332] (5) Constructing a multimodal convolutional neural network optimization model:

[0333] like Figure 2 and Figure 3 As shown, firstly, a multi-channel convolutional noise reduction feature extraction layer is designed:

[0334] Input multimodal parameter feature vector , Indicates biofilm thickness. Indicates metallic materials. Indicates the thickness of the metal film. Indicates the number of grating teeth. Indicates the spacing between the grating bumps and recesses. Indicates the wavelength of the light source. Indicates the angle of incidence. Kernel size. Convolution kernel weights can be matrixed It is a 3×3×7 tensor, the activation function is ReLU, and the bias term... It is a vector of length 3. The specific convolution calculation formula is as follows:

[0335]

[0336] Next, a multimodal parameter space-parameter correlation graph convolutional network (DCNN) is constructed:

[0337] An adaptive attention mechanism is used to assign weights to edges between nodes with different parameters, allowing for adaptive weight adjustment to identify which parameter combinations are most sensitive to SPR detection. The weights change dynamically, enabling the model to select appropriate parameter combinations under different test conditions. Next, feature transformation and convolution operations are performed, converting the features of each node into high-level embedded features through a shared convolution kernel, and then performing cross-node convolution operations on these features.

[0338] Finally, an attention fusion mechanism—multi-layer feature fusion—is implemented to ensure that the model's contribution to each feature layer can be adaptively optimized before the final output. The formula is as follows:

[0339]

[0340] (6) Loss function and output layer design:

[0341]

[0342] In the formula, and The matrices are 10×7 and 1×7 respectively; Z is the feature representation after multi-layer fusion.

[0343] Simultaneously, the mean squared error loss function is used to measure the deviation between the model's predicted parameter values ​​and the actual optimal values. By minimizing this loss, the model can continuously adjust the parameter prediction results during training to gradually approach the optimal sensor design.

[0344] The formula for calculating the mean squared error (MSE) is as follows:

[0345]

[0346] Used to calculate the true value Compared with the predicted value The smaller the squared error between the two sides, the smaller the loss and the better the prediction result.

[0347] Simultaneously, it can be combined with Mean Absolute Error (MAE) and The coefficient of determination is used to further optimize model performance. It is used to measure the proportion of variance explained by the model, indicating the goodness of fit of the model. The closer the value is to 1, the better the model fits the data.

[0348] The formula for calculating the Mean Absolute Error (MAE) is:

[0349]

[0350] The formula for calculating the coefficient of determination is:

[0351]

[0352] Simultaneously, batch normalization is applied after each convolutional or fully connected layer throughout the training process to ensure a stable distribution of output values ​​across layers. This normalizes activation values, reduces internal covariance shift, and thus improves the training speed and stability of the model. The formula for batch normalization is:

[0353]

[0354] In the formula, and These are the batch mean and variance, respectively. It is a decimal to prevent the denominator from being zero; normalization ensures the model's learning ability is reliable. It is the actual value. This is a predicted value.

[0355] Furthermore, to prevent overfitting, an exponential decay mechanism is used to dynamically adjust the learning rate. Training stops when the model's validation set loss no longer decreases significantly, thus preventing overtraining.

[0356]

[0357] In the formula This is the initial learning rate, which is 0.01 in this example; is the decay rate, which is -0.0005 in this example; t is the number of training steps.

[0358] The model performance was evaluated using 10-fold cross-validation. Through multiple feedback training iterations, the model can automatically calibrate the coupling effect between the metal film thickness and the incident angle to ensure maximum sensitivity for each analyte. The process is as follows:

[0359]

[0360] In the formula, This represents the mean squared error of the i-th fold, and the average value of the 10 folds is taken to obtain the stability and generalization ability of the model.

[0361] The trained DCNN neural network is tested using test set samples to realize the design of Kretschmann grating SPR sensor, and the sensor structure design for each analyte is obtained.

[0362] To verify the accuracy of the Kretschmann grating-type SPR sensor design in this invention, five groups of analytes were tested to design Kretschmann grating-type SPR sensors. The experimental results are as follows: Figure 4As shown, the Kretschmann grating-type SPR sensor design method established in this invention maintains an accuracy rate of over 98% for sensor structure design of unknown analytes, achieving high accuracy while ensuring stability, demonstrating good performance. This indicates that the Kretschmann grating-type SPR sensor design method established in this invention is effective, providing a better method for the detection of unknown molecules and possessing certain practicality.

Claims

1. A method for optimizing the parameters of an SPR grating-type sensor based on FPCL-DCNN, characterized in that, The specific steps are as follows: Step 1: An unknown gas or liquid is used as the analyte molecule. Calculate the elements involved in the reaction between the surface of the bio-based membrane and the metal layer material of the SPR sensor; The elements include: active groups on the surface of the bio-based membrane of the analyte molecule X. After chemical bonding with the metal layer material, the modification concentration of the analyte molecule and the remaining concentration of active groups B on the base film surface. ; Analytical molecules in gaseous or liquid form The number of sites n on the base film surface after physical adsorption and the degree of reaction on the base film surface, i.e., surface coverage. ; After chemical and physical modification, the test molecule Residual concentration in the medium channel ; and the comprehensive applicability parameter S of the biofilm; Step 2: Determine whether each element exceeds its respective set threshold. If so, proceed to Step 3; otherwise, it indicates that the molecule to be tested... The surface of the bio-based membrane did not react with the metal layer material of the SPR sensor. After replacing the metal layer material of the sensor, the values ​​of the reacting elements were recalculated. Step 3: Set the molecule to be tested The initial parameters of the metal layer material of the SPR sensor were determined and optimized using the DCNN optimization model to improve the response performance. Initialization parameters include initial biofilm thickness The reflectivity R of the reactive metal film material and thickness, and the number of teeth in the grating parameters. and the gap between the concave and convex surfaces and the wavelength of the incident light source With the angle of incidence : Step 4: Design and optimize the multi-channel convolutional noise reduction feature extraction layer of the DCNN model to extract multi-channel features from the initial parameters; Input features: biofilm thickness, metal film material, metal film thickness, number of grating teeth, concave-convex spacing, incident angle, and incident light wavelength. Convolution operations are applied, gradually increasing the feature dimension through multiple layers of 1D convolutions, and combined with a noise reduction layer. The output encoded features are as follows: In the formula, It is the first The convolutional output of the layer, It is the first The convolutional kernel weights of the layer, Represents a local region of the input data. It is the first Layer bias terms, It is an activation function. It is the number of convolution kernels; Step 5: Construct a space-parameter relationship graph using the initial parameters; Each input parameter—biofilm thickness, metal film material, metal film thickness, number of grating teeth, concave-convex spacing, incident angle, and incident light wavelength—is defined as a node in the graph. The edge weights between different nodes represent the correlation strength between the parameters. Step 6: Use an adaptive attention mechanism to assign weights to the edges between nodes with different parameters in the association graph, and perform adaptive weight adjustment; The adaptive attention mechanism includes a local attention mechanism and a global attention mechanism; First, the local attention mechanism pools the multi-channel feature map output by the extraction layer to obtain global statistical information for each channel; Then, a fully connected layer is used to compress global statistics into channel-weighted coefficients, and then the contribution of different channels is recalibrated through attention weights; Channel weighting coefficient The formula is as follows: In the formula, W is the weight matrix of the fully connected layer, and pool is the global pooling operation. It is the total number of all channels; Finally, the global attention mechanism utilizes pooling and channel weighting coefficients to fuse features from multi-layer DCNN and convolutional layers, obtaining a multi-dimensional feature representation while suppressing noise or unimportant information; the formula is as follows: In the formula, L is the number of pooling layers; Indicates the first Feature representation of layer nodes; Step 7: Optimize the model. The output layer of the DCNN is designed as a linear activation layer, outputting the optimal values ​​of each parameter. ; The formula for the linear combination of output is: In the formula, and These are the weight matrix and bias of the output layer, respectively, and Z is the feature representation after multi-layer fusion. Step 8: Use the loss function that minimizes the mean squared error (MSE), the mean absolute error (MAE), and... The coefficient of determination measures the optimal value. The deviation from the actual optimal value is used to adjust the parameters and predict the results.

2. The method as described in claim 1, characterized in that, In step one, the modification concentration The calculation is as follows: The molecule to be tested The initial concentration, The equilibrium constant is the reaction equilibrium constant. Indicates time; Remaining concentration for: In the formula, Active groups on the surface of bio-based membranes The initial concentration; The number of sites n is calculated as follows: ; The number of sites n in the liquid is calculated as follows: N is the molecule to be tested. The total number of physical adsorption sites on the base film surface. gaseous molecule to be tested The partial pressure, where C is the liquid molecule to be tested. The equilibrium concentration, where K is the adsorption equilibrium constant: ; The molecule to be tested The concentration of products generated by the reaction of active group B on the base film surface with the metal layer material; Surface coverage : ; Surface coverage : Remaining concentration : In the formula, Let Avogadro's constant be 1. It is the volume of the entire medium channel of the sensor; Comprehensive applicability parameter S of bio-based membrane: In the formula, , , These correspond to the functional group reaction matching degree F and the physicochemical reaction matching degree, respectively. The weighting parameter of the similarity and compatibility matching degree L.

3. The method as described in claim 2, characterized in that, The functional group reaction matching degree The calculation is as follows: In the formula, It is the equilibrium constant of functional group reactions. To estimate the molecule to be tested The concentration of functional groups can be reflected in the middle. This represents the concentration of functional groups on the surface of the biofilm. It is a normalization constant, which makes The value is between 0 and 1; The number of specific functional groups contained on the surface of the biofilm; The number of functional group ligands in the analyte X that can bind to the properties of the base membrane; Physicochemical reaction matching degree The calculation is as follows: In the formula, It is the slope of the curve; The measured physical adsorption free energy; This is the median adsorption free energy. When the adsorption of the analyte between the biofilm and the membrane is observed to be at a moderate level, ; The similarity and compatibility matching degree L is calculated as follows: In the formula, Indicates the polarity of the base film. Representative of the substance to be tested polarity, This indicates the difference in polarity between the two. It is the predefined maximum polarity difference.

4. The method as described in claim 1, characterized in that, In step three, the thickness of the bio-based membrane... definition: in, It is the surface area of ​​the biofilm. These are the mass and density of the bio-based membrane, respectively. Reflectivity R is expressed as: The transmission matrix M is: The elements of the matrix represent the optical properties of each layer; Number of teeth It is the number of grooves in a grating within a certain length, and the spacing between the grooves and the convexities. There is an inverse proportional relationship, expressed as: Concave-convex spacing : For diffraction order, The wavelength of the light source, These are the incident angle and diffraction angle corresponding to this light source, respectively.

5. The method as described in claim 1, characterized in that, In step five, the calculation formula for DCNN is as follows: In the formula, It involves adding a self-connected adjacency matrix. It is an adjacency matrix The degree matrix, Indicates the first Feature representation of layer nodes, It is the first Layer-learnable weight matrix.

6. The method as described in claim 1, characterized in that, In step eight, the formula for calculating the mean square error (MSE) is as follows: The molecule to be tested The initial parameters of the bio-based membrane, the initial parameters of the metal layer material of the SPR sensor, and the actual values ​​of each element involved in the reaction between the two. The molecule to be tested Initial parameters of the bio-based membrane, initial parameters of the metal layer material of the SPR sensor, and predicted values ​​of each element involved in the reaction between the two. It is the sample size; The formula for calculating the Mean Absolute Error (MAE) is: The formula for calculating the coefficient of determination is: The molecule to be tested The initial parameters of the bio-based membrane, the initial parameters of the metal layer material of the SPR sensor, and the average values ​​of each element in the reaction.

Images