Colloidal gold multi-index synchronous detection system for AI multi-task scheduling

Abstract

The invention relates to the technical field of colloidal gold detection, and discloses a colloidal gold multi-index synchronous detection system for AI multi-task scheduling. The system comprises a spectral feature decoupling module which separates overlapped spectral responses based on a graph convolution network and generates a spectral fingerprint spectrum; the multi-index quantization module analyzes the nonlinear mapping relation through a variational auto-encoder to generate a quantization decision vector; the fluid dynamic modeling module is combined with a Navier-Stokes equation to invert a sample diffusion path; the signal drift suppression module suppresses background interference by applying a generative adversarial network; the task scheduling engine module adopts a Monte Carlo tree search strategy to allocate computing resources; the cross interference compensation module generates a compensation coefficient matrix by using a tensor decomposition algorithm; and the feedback module controls the micro-valve array to optimize the detection synchronism. All the modules cooperate to achieve multi-index synchronous detection, the detection precision, efficiency and anti-interference capability are improved, and the system is suitable for the fields of medical diagnosis, food safety detection and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of colloidal gold detection, specifically to an AI multi-task scheduling colloidal gold multi-index synchronous detection system. BACKGROUND

[0002] In the field of biomedical detection and analysis, colloidal gold detection technology is widely used due to its simple operation and low cost. However, traditional colloidal gold detection systems face many problems when dealing with multi-index synchronous detection.

[0003] From the perspective of spectral analysis, traditional systems have difficulty effectively processing the overlapping spectral response in the multi-band reflectance spectrum of colloidal gold test strips. Since the spectral characteristics of different indicators may overlap, traditional methods cannot accurately separate these overlapping signals, resulting in inaccurate determination of key parameters such as characteristic peak shift, half-peak width, and signal-to-noise ratio, which in turn affects the detection accuracy of the target analyte.

[0004] In terms of multi-index quantification, traditional systems have limited ability to analyze the non-linear mapping relationship between target analyte concentration and signal intensity. In colloidal gold detection, the antigen-antibody binding kinetics process is complex, and traditional methods cannot accurately construct a theoretical response surface or efficiently fit the measured signal shift, resulting in inaccurate quantization decision vectors such as concentration gradient, confidence interval, and cross-reactivity coefficient, which cannot meet the accuracy requirements of multi-index synchronous detection.

[0005] In terms of fluid dynamic modeling, traditional systems lack real-time acquisition and effective analysis of chromatography membrane capillary flow rate data. They cannot accurately invert the sample diffusion path using the Navier-Stokes equation, making it difficult to generate reliable chromatography front rate and diffusion resistance factor, which affects the stability and consistency of the detection process.

[0006] In the signal processing process, traditional systems lack sufficient signal drift suppression capability. Environmental noise and other factors can interfere with the detection signal, and traditional methods cannot simulate environmental noise distribution and effectively suppress background interference, making it difficult to generate baseline drift correction and peak symmetry indicators, resulting in a decrease in the reliability of the detection results.

[0007] In terms of task scheduling, traditional systems lack efficient resource allocation strategies when dealing with multiple tasks. They cannot construct a reasonable multi-task priority evaluation matrix or dynamically allocate computing resources using effective algorithms, resulting in suboptimal parallel detection instructions such as channel activation sequence and sampling frequency parameters, which leads to low detection efficiency and cannot meet the timeliness requirements of multi-index synchronous detection.

[0008] In addition, the traditional system also has obvious defects in cross interference compensation and feedback control. It cannot fuse multi-dimensional data to extract common interference patterns, and the generated compensation coefficient matrix, such as temperature compensation factor and batch difference correction, is not accurate enough. At the same time, it cannot adjust the residence time of the reaction chamber according to real-time data to optimize the detection synchronism, which affects the overall detection effect.

[0009] With the increasing demand of biomedical detection for multi-index, high-precision and high-efficiency detection, these deficiencies of the traditional colloidal gold detection system seriously restrict its application range and detection performance, so a new detection system that can solve the above problems is needed. SUMMARY

[0010] The purpose of the present application is to provide an AI multi-task scheduling colloidal gold multi-index synchronous detection system to solve the problems raised in the background art.

[0011] To achieve the above purpose, the present application provides the following technical solution: an AI multi-task scheduling colloidal gold multi-index synchronous detection system, the system comprising: A spectral feature decoupling module based on multi-band reflectance spectroscopy of colloidal gold test strips, using a graph convolution network to separate overlapping spectral responses, performing multi-scale feature fusion on the decoupled data, and generating a spectral fingerprint map; A multi-index quantification module based on the spectral fingerprint map, constructing an index response space through a variational autoencoder, analyzing the nonlinear mapping relationship between the target analyte concentration and the signal intensity, and generating a quantization decision vector; A fluid dynamic modeling module based on the quantization decision vector, real-time acquisition of chromatography membrane capillary flow rate data, inversion of sample diffusion path combined with Navier-Stokes equation, and generation of fluid dynamics parameters; A signal drift suppression module based on fluid dynamics parameters, using a generative adversarial network to simulate environmental noise distribution, suppressing background interference through attention mechanism weighted feature channels, and generating corrected detection signals; A task scheduling engine module based on the corrected detection signals, constructing a multi-task priority evaluation matrix, dynamically allocating computing resources using a Monte Carlo tree search strategy, and generating parallel detection instructions; A cross interference compensation module based on the spectral fingerprint map, the quantization decision vector, the fluid dynamics parameters, the corrected detection signals and the parallel detection instructions, using a tensor decomposition algorithm to fuse multi-dimensional data, extracting common interference patterns through non-negative matrix decomposition, and generating a compensation coefficient matrix; A feedback module based on the compensation coefficient matrix, controlling the opening and closing sequence of the micro-valve array, adjusting the residence time of the reaction chamber combined with real-time impedance monitoring data, optimizing the detection synchronism through dynamic programming algorithm, and generating feedback control instructions.

[0012] Preferably, the spectral fingerprint specifically comprises characteristic peak shift amount, half-peak width and signal-to-noise ratio parameters, the quantitative decision vector includes concentration gradient value, confidence interval and cross-reaction coefficient, the hydrodynamic parameter specifically refers to chromatographic front rate and diffusion resistance factor, the corrected detection signal includes baseline drift correction amount and peak symmetry index, the parallel detection instruction contains channel activation sequence and sampling frequency parameters, the compensation coefficient matrix includes temperature compensation factor and batch difference correction amount, and the feedback control instruction specifically is pressure pulse width modulation parameter and flow control threshold.

[0013] Preferably, the spectral feature decoupling module comprises a spectral acquisition submodule, a feature extraction submodule and a graph generation submodule. The spectral acquisition submodule captures the reflectance spectrum of the test strip based on a hyperspectral imager, separates the visible light and near-infrared waveband using a CMOS linear array image sensor technology, eliminates environmental light interference by a photon counting method, and generates an original spectral data set. The feature extraction submodule applies a wavelet packet transform algorithm to decompose spectral band components based on the original spectral data set, calculates the energy entropy features of each sub-band, and constructs a multi-resolution feature space. The graph generation submodule aggregates adjacent waveband correlation features using a graph neural network based on the multi-resolution feature space, divides the feature response region through a spectral clustering algorithm, and generates a spectral fingerprint graph.

[0014] Preferably, the multi-index quantification module comprises a response modeling submodule, a nonlinear mapping submodule and a decision generation submodule. The response modeling submodule constructs an antigen-antibody binding kinetics differential equation based on the affinity characteristics of gold-labeled antibodies, solves the reaction equilibrium point by the Runge-Kutta method, and generates a theoretical response surface. The nonlinear mapping submodule applies a radial basis function network to fit the measured signal offset based on the theoretical response surface, and optimizes the network weight parameters using the conjugate gradient method. The decision generation submodule calculates the posterior probability distribution of the target object based on the optimized mapping relationship through Bayesian inference, and generates a quantitative decision vector.

[0015] Preferably, the fluid dynamic modeling module comprises a flow rate monitoring submodule, a path reconstruction submodule and a parameter generation submodule. The flow rate monitoring submodule scans the surface of the chromatography membrane using a laser Doppler velocimeter, calculates the displacement vector of the tracer particles through a cross-correlation algorithm, and generates a spatiotemporal flow rate field. The path reconstruction submodule applies a level set method to track the deformation process of the liquid flow front based on the spatiotemporal flow rate field, and constructs a three-dimensional capillary network topology structure. The parameter generation submodule generates fluid dynamics parameters based on the capillary network topology and adopts the lattice Boltzmann method to simulate the permeability of heterogeneous media.

[0016] Preferably, the signal drift suppression module includes a noise generation submodule, a feature weighting submodule, and a signal output submodule. The noise generation submodule constructs a conditional generative adversarial network model based on the device temperature drift characteristics and generates environmental noise simulation data through a gradient penalty mechanism. The feature weighting submodule calculates feature map importance weights using a channel attention mechanism based on noise simulation data and filters low-correlation feature channels using a gating unit. The signal output submodule extracts effective signal regions through adaptive threshold segmentation based on weighted feature maps and generates corrected detection signals.

[0017] Preferably, the task scheduling engine module includes an evaluation matrix construction submodule, a resource allocation submodule, and an instruction generation submodule. The evaluation matrix construction submodule defines a dual-dimension evaluation index of computational complexity and timeliness based on the detection task deadline constraint and generates a task priority matrix. The resource allocation submodule applies the Hungarian algorithm to solve an optimal task-resource matching scheme based on the task priority matrix and verifies the feasibility of the scheme through time window constraints. The instruction generation submodule generates parallel detection instructions by triggering detection channel enable signals using an event-driven architecture based on the matching scheme.

[0018] Preferably, the cross-interference compensation module includes a tensor construction submodule, a mode decomposition submodule, and a matrix generation submodule. The tensor construction submodule organizes spectral, temporal, and spatial dimensions into a third-order tensor structure based on multi-batch historical detection data. The mode decomposition submodule applies the CP decomposition algorithm to extract common interference basis vectors based on the third-order tensor structure and optimizes decomposition residuals through the alternating least squares method. The matrix generation submodule adopts sparse coding to reconstruct current detection interference components based on interference basis vectors and generates compensation coefficient matrices.

[0019] Preferably, the feedback module includes a sequence planning submodule, a parameter adjustment submodule, and an instruction generation submodule. The sequence planning submodule applies Petri net modeling to valve state transition logic based on the reaction chamber topology and generates optimal action sequences. The parameter adjustment submodule adopts model predictive control algorithm to rollingly optimize the retention time set value based on real-time impedance phase angle data. The instruction generation submodule generates feedback control instructions by converting the voltage control quantity through a pulse width modulation encoder based on a set value.

[0020] Preferably, the nonlinear mapping submodule adopts an incremental training strategy, updates network hidden layer nodes in real time through an online sequence extreme learning machine, and combines a regularization parameter to prevent overfitting.

[0021] Compared with the prior art, the AI multi-task scheduling colloidal gold multi-index synchronous detection system has the following beneficial effects: The spectral feature decoupling module uses a graph convolution network to separate overlapping spectral responses and perform multi-scale feature fusion to generate a spectral fingerprint containing key parameters such as feature peak displacement, effectively improving the accuracy of spectral analysis and laying an accurate data foundation for subsequent detection. The multi-index quantification module constructs an index response space with the help of a variational autoencoder to analyze the nonlinear mapping relationship between the concentration of the target analyte and the signal intensity, and generates an accurate quantification decision vector, making the detection of the target concentration more accurate and reliable.

[0022] The fluid dynamic modeling module collects chromatography membrane capillary flow rate data in real time, combines the Navier-Stokes equation to invert the sample diffusion path, and generates fluid dynamics parameters, which can accurately grasp the sample flow state and provide a guarantee for the stability of the detection process. The signal drift suppression module applies a generative adversarial network to simulate environmental noise distribution, suppresses background interference through an attention mechanism to weight feature channels, and generates corrected detection signals, effectively improving the quality and reliability of the signals and reducing the influence of environmental factors on the detection results.

[0023] The task scheduling engine module constructs a multi-task priority evaluation matrix, dynamically allocates computing resources using a Monte Carlo tree search strategy, generates parallel detection instructions, optimally configures computing resources, greatly improves the efficiency of multi-task processing, and meets the timeliness requirements of multi-index synchronous detection. The cross-interference compensation module applies a tensor decomposition algorithm to fuse multi-dimensional data, extracts common interference patterns through non-negative matrix decomposition, generates a compensation coefficient matrix, effectively compensates for cross-interference, and improves the accuracy and consistency of detection.

[0024] The feedback module controls the opening and closing sequence of the micro-valve array, adjusts the residence time of the reaction chamber in combination with real-time impedance monitoring data, optimizes the detection synchronization through a dynamic programming algorithm, generates feedback control instructions, and realizes precise regulation of the detection process, further improving the overall effect of detection. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The working principle diagram of the AI multi-task scheduling colloidal gold multi-index synchronous detection system described in the application; Figure 2 The working principle diagram of the signal drift suppression module; Figure 3 Design diagram of the cross interference compensation module; Figure 4 Control schematic of the feedback module. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the protection scope of the present application.

[0027] Please refer to Figures 1-4 The AI multi-task scheduling colloidal gold multi-index synchronous detection system provided by the present application comprises: A spectral feature decoupling module, based on multi-band reflection spectrum of a colloidal gold test strip, adopts a graph convolution network to separate overlapping spectral responses, performs multi-scale feature fusion on decoupled data, and generates a spectral fingerprint spectrum; A multi-index quantification module, based on the spectral fingerprint spectrum, constructs an index response space through a variational autoencoder, analyzes a nonlinear mapping relationship between a target analyte concentration and a signal intensity, and generates a quantification decision vector; A fluid dynamic modeling module, based on the quantification decision vector, collects chromatography membrane capillary flow rate data in real time, inverses a sample diffusion path in combination with a Navier-Stokes equation, and generates fluid dynamics parameters; A signal drift suppression module, based on the fluid dynamics parameters, applies a generative adversarial network to simulate environmental noise distribution, suppresses background interference through an attention mechanism to weight feature channels, and generates a corrected detection signal; A task scheduling engine module, based on the corrected detection signal, constructs a multi-task priority evaluation matrix, dynamically allocates computing resources by adopting a Monte Carlo tree search strategy, and generates parallel detection instructions; A cross interference compensation module, based on the spectral fingerprint spectrum, the quantification decision vector, the fluid dynamics parameters, the corrected detection signal and the parallel detection instructions, applies a tensor decomposition algorithm to fuse multi-dimensional data, extracts a common interference mode through non-negative matrix decomposition, and generates a compensation coefficient matrix; A feedback module, based on the compensation coefficient matrix, controls opening and closing sequences of a micro-valve array, adjusts a reaction chamber residence time in combination with real-time impedance monitoring data, optimizes detection synchronism through a dynamic programming algorithm, and generates feedback control instructions.

[0028] The present application will be further described below in combination with Examples 1 to 5: Example 1 The embodiment describes in detail the definition and role of parameters involved in each module of the system. The displacement amount of the characteristic peak included in the spectral fingerprint is the change in the position of the spectral characteristic peak on the wavelength axis. It can directly reflect the influence of the change in the molecular structure or environment of the substance on the spectrum. For example, when the particle size of the colloidal gold particles or the refractive index of the surrounding medium changes, the position of the characteristic peak will move accordingly. The half-peak width represents the width of the characteristic peak at half height, which is related to the purity of the substance, the uniformity of particle distribution, etc. A wider half-peak width may mean that there are colloidal gold particles of multiple particle sizes or impurities in the sample. The signal-to-noise ratio parameter is used to measure the proportion of effective components and noise in the signal. Spectral data with high signal-to-noise ratio can more clearly show the characteristics of the substance, providing a reliable basis for subsequent analysis.

[0029] The concentration gradient value in the quantitative decision vector is obtained by analyzing the rate of change of the target analyte concentration in space or time. It can help to judge the distribution of the target substance in the detection sample. For example, in the detection of blood samples, the concentration gradient value can reflect the distribution difference of the target analyte in different regions. The confidence interval is the uncertainty range of the estimation of the target substance concentration, which is calculated based on statistical methods and provides a reliability index for the concentration measurement result. For example, when the confidence interval is small, the estimation of the concentration is more accurate. The cross-reaction coefficient is used to represent the degree of interference of other substances on the detection of the target analyte during the detection process. The larger the coefficient, the more obvious the cross-reaction, and the worse the specificity of the detection. In multi-index synchronous detection, accurately determining the cross-reaction coefficient is crucial for eliminating interference and improving detection accuracy.

[0030] The chromatographic front rate in the fluid dynamics parameter is the speed at which the sample advances in the chromatographic membrane. It is affected by factors such as the material and pore size of the chromatographic membrane, the viscosity of the sample, etc. Different chromatographic front rates will directly affect the reaction time and binding efficiency of the sample and the colloidal gold label. The diffusion resistance factor reflects the degree of hindrance to the sample diffusion in the chromatographic membrane. Its size is closely related to the microstructure of the chromatographic membrane. A larger diffusion resistance factor may cause slow sample diffusion, affecting the timeliness and accuracy of the detection.

[0031] The baseline drift correction amount in the corrected detection signal is to eliminate the baseline shift caused by factors such as environmental temperature change and instrument stability during the detection process. The stability of the baseline directly affects the recognition and quantitative analysis of the signal peak. The peak symmetry index is used to measure the symmetry degree of the signal peak. An ideal symmetric peak shape helps to more accurately determine the position and area of the peak, thereby improving the accuracy of concentration quantification. If the peak shape is asymmetric, it may be caused by sample matrix interference, abnormal reaction kinetics, etc.

[0032] The channel activation sequence included in the parallel detection instruction specifies the order of opening and closing of each detection channel. In multi-index synchronous detection, a reasonable channel activation sequence can avoid mutual interference between different index detections, improving the parallelism and efficiency of detection. The sampling frequency parameter determines the number of times the signal is sampled during the detection process. A suitable sampling frequency can ensure that sufficient signal information is obtained without increasing the amount of data processing.

[0033] The temperature compensation factor in the compensation coefficient matrix is used to eliminate the influence of temperature changes on the detection results. Temperature fluctuations can affect the activity of colloidal gold markers, the rate of antigen-antibody binding reactions, and other factors. The batch difference correction amount is used to compensate for performance differences between different batches of detection reagents or instruments. Due to minor changes in the reagent production process or different instrument calibrations, different batches of detection may produce systematic deviations. The batch difference correction amount can correct such deviations.

[0034] The feedback control instruction specifies the pressure pulse width modulation parameters, which are used to control the size and action time of the pressure in the system, thereby adjusting the flow of the sample in the reaction chamber. The flow control threshold sets the upper and lower limits of the sample flow, ensuring that the flow is within a reasonable range to maintain stable reaction conditions. These parameters need to be dynamically adjusted according to specific detection scenarios and requirements in practical applications. For example, when detecting different types of samples, the chromatography front rate and diffusion resistance factor may change due to differences in sample viscosity and composition. In this case, the fluid dynamics parameters need to be adjusted accordingly. When the detection environment temperature changes significantly, the temperature compensation factor needs to be changed to ensure the accuracy of the detection results.

[0035] Example 2: This embodiment details the specific implementation of the spectral feature decoupling module. Through the cooperative work of multiple sub-modules, the module achieves efficient decoupling and feature extraction of the reflectance spectrum of the colloidal gold test strip. The spectral acquisition sub-module, as the starting link of the module, captures the reflectance spectrum of the test strip based on a hyperspectral imager. The hyperspectral imager can cover a wide spectral range, providing rich data basis for subsequent analysis by acquiring continuous spectral information from visible light to near-infrared bands. During spectral acquisition, CMOS line array image sensor technology is used to separate light of different wavelengths, ensuring that spectral information of each band can be independently acquired. To eliminate the interference of ambient light, the sub-module processes the light signal through photon counting method, which is a high-sensitivity detection method that can convert weak light signals into discrete photon counts, effectively suppressing background noise and improving the signal-to-noise ratio of the original spectral data set, making the collected spectral data more accurately reflect the reflectance characteristics of the test strip.

[0036] The feature extraction submodule further processes the original spectral dataset generated by the spectral acquisition submodule. This submodule applies the wavelet packet transform algorithm to decompose the spectral band components. Wavelet packet transform is a more refined signal decomposition method than traditional wavelet transform, which not only decomposes the low-frequency part of the signal but also decomposes the high-frequency part, enabling the spectral data to be decomposed into multiple different frequency bands, each corresponding to different frequency components. After decomposition, the energy entropy features of each sub-band are calculated. Energy entropy can reflect the energy distribution in each sub-band, representing the complexity and uncertainty of spectral data in different frequency bands. By extracting these energy entropy features, a multi-resolution feature space is constructed, which contains the feature information of the spectrum at different scales and resolutions, providing more abundant feature descriptions for subsequent map generation.

[0037] The map generation submodule generates the spectral fingerprint map based on the multi-resolution feature space. This submodule uses a graph neural network to aggregate the correlation features of adjacent bands. Graph neural network is a neural network specifically designed for processing graph-structured data. In spectral analysis, each band is considered as a node in the graph, and the correlation between adjacent bands is considered as an edge. Through the graph neural network, the spatial dependence and local features between bands can be effectively captured, thereby realizing the aggregation of adjacent band features. After aggregating the features, the spectral clustering algorithm is used to divide the feature response regions. Spectral clustering is a clustering method based on graph theory, which uses the eigenvalues and eigenvectors of the similarity matrix of sample data for clustering, and can cluster bands with similar features into a class, thereby dividing different feature response regions. Each feature response region corresponds to a specific substance or reaction feature, and the final generated spectral fingerprint map can intuitively display these feature response regions and their corresponding feature parameters, such as characteristic peak displacement, half-peak width, etc., providing an important basis for subsequent multi-index quantitative analysis.

[0038] In practical applications, the three sub-modules are closely connected and interact with each other. The quality of the original spectral data obtained by the spectral acquisition sub-module directly affects the processing effect of the feature extraction sub-module. If there is a lot of noise in the original data or the band separation is not accurate, it will lead to inaccurate extraction of features, and then affect the quality of the generated atlas. The richness and accuracy of the multi-resolution feature space constructed by the feature extraction sub-module determine whether the atlas generation sub-module can accurately divide the feature response area and generate reliable spectral fingerprint atlas. For example, when detecting different types of colloidal gold test strips, the corresponding detection targets and reaction mechanisms are different, and the characteristics of the reflected spectrum will also be different. For test strips that detect specific antigens, the position and intensity of the spectral feature peaks will change with the change of antigen concentration. The spectral acquisition sub-module needs to adjust the acquisition parameters of the hyperspectral imager, such as wavelength range, exposure time, etc., according to the spectral characteristics, to ensure that the original spectral data containing key features is obtained. The feature extraction sub-module needs to adjust the decomposition level and parameters of the wavelet packet transform according to these features, in order to extract the energy entropy features that best reflect the change of antigen concentration. The atlas generation sub-module then accurately divides the feature response area related to the antigen concentration according to these features, and generates the corresponding spectral fingerprint atlas.

[0039] In addition, various factors in the actual detection environment also need to be considered in the processing of each sub-module of this module. For example, changes in environmental temperature may cause changes in the physical properties and chemical reactions of the test strip, thereby affecting the characteristics of the reflected spectrum. In the spectral acquisition sub-module, certain temperature control measures need to be taken or the influence of temperature needs to be considered in data processing to ensure the stability of the collected spectral data. In the feature extraction and atlas generation process, spectral shifts caused by temperature changes also need to be corrected to ensure that the generated spectral fingerprint atlas can accurately reflect the information of the target analyte.

[0040] Through the sequential processing of the three sub-modules of the spectral feature decoupling module, the conversion from original spectral data to spectral fingerprint atlas is realized, providing key spectral feature information for the entire detection system. This module not only effectively separates overlapping spectral responses, but also captures subtle changes in the spectrum through multi-scale feature fusion, thereby improving the accuracy and reliability of the detection system for multi-index synchronous detection. In multi-index synchronous detection, the spectral features corresponding to different indicators may overlap or interfere with each other. Through accurate decoupling and feature extraction, this module can distinguish the spectral features of different indicators.

[0041] Example 3: The embodiment describes the implementation of the multi-index quantification module in detail. The module realizes the accurate quantification of the target analyte concentration through the collaborative work of three sub-modules of response modeling, nonlinear mapping and decision making. The response modeling sub-module constructs the differential equation of antigen-antibody binding kinetics based on the affinity characteristics of gold-labeled antibodies. The gold-labeled antibodies are connected by physical adsorption or chemical combination of colloidal gold particles and antibodies, and the affinity characteristics determine the strength and specificity of the antigen-antibody binding reaction. In the antigen-antibody binding process, the reaction kinetics follows certain rules and can be described by a differential equation. Let the antigen concentration be , the antibody concentration be , and the concentration of the antigen-antibody complex formed by binding be , the kinetic differential equation can be expressed as:

[0042] wherein, represents the reaction time, is the positive reaction rate constant, describing the rate of antigen binding with antibody to form a complex; is the reverse reaction rate constant, describing the rate of complex dissociation into antigen and antibody. The equation reflects the relationship between the complex concentration and time during the reaction process, and the reaction equilibrium point can be obtained by solving the equation.

[0043] In order to solve the reaction equilibrium point of the above differential equation, the response modeling sub-module adopts the Runge-Kutta method. The Runge-Kutta method is a numerical method for solving ordinary differential equations, which has high precision and stability. By setting appropriate initial conditions and step size, the Runge-Kutta method can iteratively calculate the complex concentration at different time points. When the rate of change of the complex concentration tends to zero, it is considered to have reached the reaction equilibrium point, and the theoretical response surface can be generated at this time. The theoretical response surface describes the relationship between the target analyte concentration and the complex concentration under ideal conditions, providing a theoretical basis for subsequent nonlinear mapping.

[0044] The nonlinear mapping sub-module applies the radial basis function network to fit the measured signal offset based on the theoretical response surface. The radial basis function network is a kind of feedforward neural network, which consists of input layer, hidden layer and output layer. The activation function of the hidden layer nodes is radial basis function, usually Gaussian function. Let the input be the feature vector of the measured signal , the center of the th node of the hidden layer be , and the width be , then the output of the hidden layer node is:

[0045] wherein, is the output of the th node of the hidden layer the output of a node, represents the input vector the Euclidean distance to the center , represents the natural exponential function. The output of the output layer is a linear combination of the outputs of the hidden layer nodes, i.e.,

[0046] where represents the fitting result of the radial basis function network for the offset between the measured signal and the theoretical response surface, is the weight of the hidden layer node to the output layer, is the bias, is the number of hidden layer nodes.

[0047] In practical applications, the measured signal will have an offset from the theoretical response surface due to various factors such as sample matrix interference, instrument noise, etc. The nonlinear mapping submodule fits this offset through a radial basis function network. A large amount of measured data and corresponding theoretical response values are collected to form a training data set; the conjugate gradient method is used to optimize the network's weight parameters , center , and width . The conjugate gradient method is an optimization algorithm that searches for the optimal parameter combination through iterative calculations to minimize the error between the network's output and the measured signal offset. By continuously adjusting these parameters, the radial basis function network can accurately fit the nonlinear relationship between the measured signal and the theoretical response, improving the model's fitting accuracy.

[0048] The decision generation submodule generates a quantitative decision vector based on the optimized mapping relationship by calculating the posterior probability distribution of the target object through Bayesian inference. Bayesian inference is a statistical inference method based on Bayes' theorem, which combines prior knowledge and observed data to update the probability estimate of the target object concentration. Let the target object concentration be , and the measured signal be , according to Bayes' theorem, the posterior probability distribution is:

[0049] where is the prior probability distribution of the target object concentration, reflecting the understanding of the target object concentration before detection; is the likelihood function, representing the probability of observing the measured signal when the target object concentration is ; is the evidence factor, which is a normalization constant ensuring that the integral of the posterior probability distribution equals 1.

[0050] In actual calculation, the prior probability distribution can be set according to historical detection data or domain knowledge, and the likelihood function can be obtained through the optimized mapping relationship. Through calculation of the posterior probability distribution , various statistics of the target concentration, such as mean, variance, confidence interval, etc., can be obtained, which constitute the quantitative decision vector. The quantitative decision vector includes concentration gradient value, confidence interval, and cross-reaction coefficient, etc. parameters, wherein the concentration gradient value reflects the trend of the target concentration, the confidence interval provides the uncertainty range of the concentration estimation, and the cross-reaction coefficient represents the interference degree of other substances on the target detection.

[0051] In the multi-index synchronous detection process, the three sub-modules are mutually related and influenced. The accuracy of the theoretical response surface constructed by the response modeling sub-module directly affects the fitting effect of the nonlinear mapping sub-module. If there is a deviation between the theoretical response surface and the actual reaction process, it will lead to an increase in the fitting error of the radial basis function network, and further affect the posterior probability calculation of the decision generation sub-module. The fitting accuracy of the nonlinear mapping sub-module determines whether the decision generation sub-module can accurately calculate the posterior probability distribution of the target and generate reliable quantitative decision vectors. For example, when detecting different target analytes, the affinity characteristics of the gold-labeled antibody are different, and the response modeling sub-module needs to construct the corresponding kinetic differential equation according to the specific antigen-antibody system, and adjust the values of the forward reaction rate constant and the reverse reaction rate constant . The nonlinear mapping sub-module needs to adjust the structure and parameters of the radial basis function network according to the characteristics of the measured signal, so as to adapt to different nonlinear relationships. The decision generation sub-module calculates the accurate posterior probability distribution according to the optimized mapping relationship and prior knowledge, and generates the quantitative decision vector containing the concentration gradient value, confidence interval, and cross-reaction coefficient, etc. parameters.

[0052] The module also needs to consider various interference factors in the actual detection environment. For example, impurities in the sample may non-specifically bind with the gold-labeled antibody, affecting the kinetics of the antigen-antibody binding reaction, leading to an increase in the deviation between the theoretical response surface and the measured signal. In this case, the nonlinear mapping sub-module needs to improve the generalization ability of the model through more training data, and the decision generation sub-module also needs to adjust the prior probability distribution accordingly, so as to reduce the influence of impurity interference on the detection result.

[0053] Through the sequential processing of the three sub-modules of the multi-index quantitative module, the whole process from the theoretical modeling of the antigen-antibody binding reaction to the nonlinear fitting of the measured signal, and then to the probability inference and quantitative decision of the target concentration is realized.

[0054] Example 4: The embodiment describes the specific implementation of the fluid dynamic modeling module in detail. The module realizes the accurate modeling of the flow characteristics of the sample in the chromatography membrane and the generation of the fluid dynamics parameters through the collaborative work of three sub-modules of flow rate monitoring, path reconstruction, and parameter generation. The flow rate monitoring sub-module scans the surface of the chromatography membrane using a laser Doppler velocimeter. The working principle of the laser Doppler velocimeter is based on the Doppler effect. When laser irradiates on the moving tracer particles, the frequency of the reflected light will shift. By measuring the frequency shift, the movement speed of the particles can be calculated. In practical applications, an appropriate amount of tracer particles will be added to the sample. These particles need to have good optical properties and not react with the components in the sample to ensure the accuracy of the measurement. When the sample flows in the chromatography membrane, the tracer particles will move with the sample. The laser Doppler velocimeter scans the surface of the chromatography membrane at a certain scanning frequency to obtain the speed information of the tracer particles at different positions. Through cross-correlation algorithm, the displacement vectors of the tracer particles at different time points are calculated, and then the space-time flow rate field is generated. The space-time flow rate field can intuitively show the flow rate distribution of the sample at different positions and times in the chromatography membrane.

[0055] The path reconstruction sub-module applies the level set method to track the deformation process of the liquid flow front based on the space-time flow rate field generated by the flow rate monitoring sub-module. The level set method is a numerical method for describing and tracking the evolution of interfaces. It represents the liquid flow front as a zero level set of a high-dimensional function. By solving partial differential equations, the change of the liquid flow front over time is simulated. In the specific implementation process, first, the movement speed of the liquid flow at different positions is determined according to the space-time flow rate field, and then these speeds are input as boundary conditions into the level set equation. Through iterative calculation, the position and shape of the liquid flow front are updated. As the calculation progresses, the level set method can accurately capture the complex deformation processes of the liquid flow front in the chromatography membrane pores, such as bending and branching, thereby constructing a three-dimensional capillary network topology structure. The topology structure describes the distribution, connection relationship, and pore size of the capillary channels in the chromatography membrane in detail, providing a geometric basis for the subsequent generation of fluid dynamics parameters.

[0056] The parameter generation submodule generates fluid dynamics parameters based on the three-dimensional capillary network topology constructed by the path reconstruction submodule. The lattice Boltzmann method is used to simulate the permeability of the heterogeneous medium. The lattice Boltzmann method is a mesoscopic-scale fluid simulation method that treats fluid as a system composed of a large number of particles that move and interact on a regular lattice. During the simulation process, the three-dimensional capillary network topology is first discretized into a lattice model, with each lattice representing a small area in the chromatography membrane. Each lattice is assigned corresponding physical properties such as porosity and pore size based on the characteristics of the topology. Then, by solving the lattice Boltzmann equation, the flow process of the fluid in the discretized model is simulated, and the flow rate of the fluid under different pressure gradients is calculated. Permeability is a physical quantity that measures the ability of a medium to allow fluid to pass through, and it can be calculated from the relationship between flow rate and pressure gradient. In a heterogeneous medium, the permeability of different regions may differ, and the lattice Boltzmann method can well simulate this heterogeneity, thus accurately calculating the permeability distribution of the entire chromatography membrane. Based on the permeability distribution and other related physical parameters such as fluid viscosity and pressure, the fluid dynamics parameters are finally generated, including the chromatography front rate and the diffusion resistance factor.

[0057] In actual detection scenarios, the coordinated work of these three submodules is particularly evident. For example, when detecting a specific marker in a blood sample, the blood sample contains multiple components with different viscosities and flow properties from pure water. The flow rate monitoring submodule needs to adjust the scanning parameters of the laser Doppler velocimeter, such as laser power and scanning frequency, according to the characteristics of the blood sample when scanning the surface of the chromatography membrane, to ensure that the motion information of the tracer particles can be accurately captured. Due to the high viscosity of the blood sample, the motion speed of the tracer particles may be slower, and the cross-correlation algorithm needs to more finely process the calculation of displacement vectors to generate an accurate spatiotemporal flow field.

[0058] The path reconstruction submodule may find that the advancement of the liquid flow front in the chromatography membrane may be hindered due to the presence of large molecules such as proteins in the blood sample, making the deformation process of the liquid flow front more complex. The level set method needs to more accurately update the position of the liquid flow front based on the speed changes reflected in the spatiotemporal flow field to construct a three-dimensional capillary network topology that conforms to the actual situation. For example, in some capillary channels with smaller pore sizes, the flow of the blood sample may exhibit hysteresis phenomena, and the level set method needs to accurately capture these details to ensure the accuracy of the topology.

[0059] The parameter generation submodule generates fluid dynamics parameters based on the constructed three-dimensional capillary network topology. Due to the influence of the non-homogeneous medium permeability on the flow of the blood sample in the chromatography membrane, the lattice Boltzmann method needs to handle the physical properties of different regions more carefully to simulate the real fluid flow process. The permeability distribution obtained by simulation, combined with the viscosity and other parameters of the blood sample, generates a lower chromatography front rate and a larger diffusion resistance factor than when detecting pure water samples. These fluid dynamics parameters can truly reflect the flow characteristics of the blood sample in the chromatography membrane, providing an important basis for subsequent fluid dynamic analysis and process optimization.

[0060] In addition, the module also needs to consider the influence of the characteristics of the chromatography membrane itself on the fluid dynamics parameters. Different materials and structures of the chromatography membrane have different porosities and pore size distributions, which will directly affect the flow speed and diffusion resistance of the sample in it. For example, a certain type of chromatography membrane has a relatively uniform pore size distribution, and the sample flows relatively stably in it. The spatiotemporal flow velocity field generated by the flow velocity monitoring submodule is also relatively uniform, the capillary channel connection in the three-dimensional capillary network topology constructed by the path reconstruction submodule is relatively regular, and the permeability distribution calculated by the parameter generation submodule is also relatively consistent. The generated chromatography front rate and diffusion resistance factor have good repeatability. Another type of chromatography membrane may have local areas with larger or smaller pore sizes, resulting in uneven flow velocity of the sample during the flow process. At this time, the spatiotemporal flow velocity field will show obvious fluctuations, and local channel expansion or narrowing will occur in the three-dimensional capillary network topology. The permeability distribution simulated by the lattice Boltzmann method will also be more complex, and the fluid dynamics parameters generated will also change accordingly.

[0061] Through the sequential processing of the three submodules of the fluid dynamic modeling module, the whole process from sample flow velocity monitoring to liquid flow path reconstruction and then to fluid dynamics parameter generation is realized. This module can accurately describe the flow characteristics of the sample in the chromatography membrane, providing key fluid dynamics parameters for subsequent signal drift suppression, cross-interference compensation, and other modules, thereby ensuring that the entire detection system can be adjusted and optimized in real time according to the flow characteristics of the sample during multi-index synchronous detection, improving the accuracy and reliability of the detection. For example, during multi-index synchronous detection, different detection reagents corresponding to different indexes may have different effects on the flow characteristics of the sample. By generating fluid dynamics parameters through the fluid dynamic modeling module, the system can adjust the control parameters of the feedback module in real time to maintain stable reaction conditions and ensure that the detection results of each index are not affected by abnormal sample flow.

[0062] Example 5: The embodiment describes in detail the cooperative working process of the signal drift suppression module, the task scheduling engine module, the cross-interference compensation module and the feedback module in the system to ensure the stability and accuracy of multi-index synchronous detection. Taking the detection of multiple tumor markers in serum as an example, the signal drift suppression module needs to process the influence of environmental temperature drift on the detection signal. The noise generation submodule constructs a conditional generative adversarial network model based on the temperature drift characteristics of the device at different temperatures. The model learns the mapping relationship between historical temperature drift data and corresponding noise, and generates noise simulation data that matches the current temperature environment. For example, when the detection environment temperature rises from 25°C to 30°C, the model will generate noise data such as baseline drift and signal fluctuation caused by simulated temperature rise. After receiving these noise simulation data, the feature weighting submodule applies a channel attention mechanism to analyze the importance weight of each channel in the spectral feature map. In the detection of tumor markers, for example, the signals of certain wavebands are easily affected by temperature and produce drift. The mechanism identifies these key channels, reduces the weight of non-key channels through a gating unit, filters out low-correlation features, and highlights the features of temperature-sensitive wavebands. The signal output submodule performs adaptive threshold segmentation on the weighted feature map, extracts the effective signal region that is not disturbed by noise, generates corrected detection signals, and ensures that subsequent analysis is based on a stable signal basis.

[0063] The task scheduling engine module plays a role in resource allocation in multi-index detection tasks. The evaluation matrix construction submodule sets time limit constraints according to the clinical needs of each tumor marker detection, such as requiring alpha-fetoprotein (AFP) detection to be completed within 15 minutes to support emergency diagnosis, and relaxing the time limit for carcinoembryonic antigen (CEA) detection to 30 minutes. At the same time, the submodule defines two-dimensional indicators of computational complexity and timeliness. AFP detection involves multi-scale feature fusion, has high computational complexity, and has strict timeliness requirements, so it is assigned a higher weight in the task priority matrix. The resource allocation submodule solves the optimal task-resource matching scheme based on this matrix using the Hungarian algorithm, assigns the high-priority AFP detection task to a processor core with faster computing speed, and verifies the feasibility of the scheme through time window constraints to ensure that the calculation is completed within the time limit. The instruction generation submodule generates parallel detection instructions based on the matching scheme using an event-driven architecture to trigger the enable signals of the corresponding detection channels, such as simultaneously activating the detection channels corresponding to AFP and CEA, setting the sampling frequency of the AFP channel to 10 times per second, and setting the sampling frequency of the CEA channel to 5 times per second, to achieve efficient scheduling of multiple tasks.

[0064] The cross-interference compensation module handles the signal overlap problem in multi-index detection. The tensor construction submodule collects historical data of multi-batch serum detection, organizes the spectral dimension (such as 450nm-700nm wavelength band), time dimension (time points in the detection process) and spatial dimension (positions of different detection channels) into a three-order tensor structure. For example, in the past 100 detections, each detection records 200 spectral wavelength points, 50 time points and 8 channel signal data, forming a 100x200x50x8 tensor (the dimensions are adjusted according to the needs in actual application). The mode decomposition submodule applies the CP decomposition algorithm to the tensor to extract common interference basis vectors, such as the detection signal of some tumor markers at 520nm wavelength. The decomposition algorithm will identify the common interference mode at this wavelength, and optimize the decomposition residual through the alternating least squares method to improve the accuracy of interference mode extraction. The matrix generation submodule reconstructs the interference component in the current detection based on the interference basis vector using sparse coding, generates a compensation coefficient matrix containing temperature compensation factors, batch difference correction amounts, etc., and compensates for the cross-interference in the current detection signal in real time.

[0065] The feedback module adjusts the fluid state of the reaction chamber according to the compensation result. The sequence planning submodule applies Petri net modeling to the microvalve state transition logic based on the topological structure of the reaction chamber to generate the optimal action sequence. For example, when it is detected that the reactions of AFP and CEA require different incubation times, the Petri net model will plan a sequence of microvalve A opening for 3 minutes and microvalve B opening for 5 minutes to ensure that the two markers complete the reaction in their respective appropriate time. The parameter adjustment submodule collects impedance phase angle data in real time, which reflects the residence time and mixing state of the sample in the reaction chamber. When the impedance phase angle shows that the residence time of the AFP detection sample is insufficient, the model predictive control algorithm is used to rollingly optimize the residence time setting value, adjust the pulse width modulation parameter to prolong the residence time of the sample in the chamber. The instruction generation submodule converts the optimized setting value into voltage control quantity through the pulse width modulation encoder to generate feedback control instructions, such as outputting a 12V voltage control to the opening degree of microvalve A to adjust the flow rate to 0.5μL / s, ensuring that the sample residence time reaches the ideal value and optimizing the detection synchronization.

[0066] The corrected signal generated by the signal drift suppression module is used as one of the bases for the task scheduling engine module to evaluate the task priority, and the resource allocation result of the task scheduling affects the data acquisition frequency of the cross interference compensation module. For example, when the AFP detection task occupies more computing resources, the cross interference compensation module will preferentially process the signal compensation of this channel. After the feedback module adjusts the fluid parameters according to the compensation coefficient matrix, the fluid dynamic modeling module needs to reacquire the capillary flow rate data to provide new fluid dynamics parameters for the signal drift suppression, forming a closed-loop control. This cooperative mechanism is particularly important in complex sample detection. When multiple tumor markers exist in serum at the same time, each module adjusts and interacts in real time to effectively suppress environmental noise, optimize task allocation, compensate for cross interference, and control fluid state, ensuring that the detection signal of each index can be accurately analyzed, and finally generating a reliable quantitative decision vector to provide multi-dimensional detection results for clinical diagnosis.

[0067] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0068] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An AI multi-task scheduling colloidal gold multi-index synchronous detection system, characterized in that: The system comprises: The spectral feature decoupling module uses a graph convolutional network to separate overlapping spectral responses based on the multi-band reflectance spectrum of colloidal gold test strips, performs multi-scale feature fusion on the decoupled data, and generates a spectral fingerprint. The multi-index quantification module, based on the spectral fingerprint, constructs the index response space through the variational autoencoder, analyzes the nonlinear mapping relationship between the target analyte concentration and signal intensity, and generates a quantitative decision vector; The fluid dynamics modeling module collects the capillary flow velocity data of the chromatographic membrane in real time based on the quantitative decision vector, combines the Navier-Stokes equation to invert the sample diffusion path, and generates fluid dynamics parameters; The signal drift suppression module uses a generative adversarial network to simulate the distribution of environmental noise based on fluid dynamics parameters, suppresses background interference through weighted feature channels using an attention mechanism, and generates a corrected detection signal. The task scheduling engine module builds a multi-task priority evaluation matrix based on the corrected detection signal, uses the Monte Carlo tree search strategy to dynamically allocate computing resources, and generates parallel detection instructions; The cross-interference compensation module is based on spectral fingerprints, quantized decision vectors, fluid dynamics parameters, corrected detection signals, and parallel detection instructions. It uses a tensor decomposition algorithm to fuse multi-dimensional data, extracts common interference patterns through non-negative matrix decomposition, and generates a compensation coefficient matrix. The feedback module controls the opening and closing sequence of the microvalve array based on the compensation coefficient matrix, adjusts the reaction chamber residence time in combination with real-time impedance monitoring data, optimizes detection synchronization through a dynamic programming algorithm, and generates feedback control instructions.

2. The AI ​​multi-task scheduling colloidal gold multi-index synchronous detection system according to claim 1 is characterized in that: The spectral fingerprint map specifically includes characteristic peak displacement, half-peak width and signal-to-noise ratio parameters; the quantitative decision vector includes concentration gradient value, confidence interval and cross-reaction coefficient; the fluid dynamics parameter specifically refers to the chromatographic front rate and diffusion resistance factor; the corrected detection signal includes baseline drift correction and peak shape symmetry index; the parallel detection instruction includes channel activation sequence and sampling frequency parameters; the compensation coefficient matrix includes temperature compensation factor and batch difference correction; the feedback control instruction specifically includes pressure pulse width modulation parameter and flow control threshold.

3. The AI ​​multi-task scheduling colloidal gold multi-index synchronous detection system according to claim 1 is characterized in that: The spectral feature decoupling module includes a spectrum acquisition submodule, a feature extraction submodule, and a spectrum generation submodule; The spectrum acquisition submodule captures the test strip reflection spectrum based on a hyperspectral imager, uses CMOS linear array image sensing technology to separate visible light and near-infrared bands, eliminates ambient light interference through photon counting, and generates an original spectrum data set; The feature extraction submodule uses the wavelet packet transform algorithm to decompose the spectral band components based on the original spectral data set, calculates the energy entropy characteristics of each subband, and constructs a multi-resolution feature space; The spectrum generation submodule is based on a multi-resolution feature space, uses a graph neural network to aggregate the associated features of adjacent bands, divides the feature response areas through a spectral clustering algorithm, and generates a spectral fingerprint spectrum.

4. The AI ​​multi-task scheduling colloidal gold multi-index synchronous detection system according to claim 1 is characterized in that: The multi-index quantification module includes a response modeling submodule, a nonlinear mapping submodule, and a decision generation submodule; The response modeling submodule constructs the antigen-antibody binding kinetic differential equation based on the affinity characteristics of the gold-labeled antibody, solves the reaction equilibrium point through the Runge-Kutta method, and generates a theoretical response surface; The nonlinear mapping submodule is based on the theoretical response surface, applies the radial basis function network to fit the measured signal offset, and uses the conjugate gradient method to optimize the network weight parameters; The decision generation submodule calculates the posterior probability distribution of the target object based on the optimized mapping relationship through Bayesian inference to generate a quantized decision vector.

5. The AI ​​multi-task scheduling colloidal gold multi-index synchronous detection system according to claim 1 is characterized in that: The fluid dynamic modeling module includes a flow rate monitoring submodule, a path reconstruction submodule, and a parameter generation submodule; The flow rate monitoring submodule uses a laser Doppler velocimeter to scan the surface of the chromatography membrane, calculates the displacement vector of the tracer particles through a cross-correlation algorithm, and generates a spatiotemporal flow rate field; The path reconstruction submodule uses the level set method to track the deformation process of the liquid flow front based on the spatiotemporal velocity field and construct a three-dimensional capillary network topology structure; The parameter generation submodule is based on the capillary network topology structure and uses the lattice Boltzmann method to simulate the permeability of heterogeneous media and generate fluid dynamics parameters.

6. The AI ​​multi-task scheduling colloidal gold multi-index synchronous detection system according to claim 1 is characterized in that: The signal drift suppression module includes a noise generation submodule, a feature weighting submodule, and a signal output submodule; The noise generation submodule builds a conditional generative adversarial network model based on the device temperature drift characteristics and generates environmental noise simulation data through a gradient penalty mechanism; The feature weighting submodule calculates the feature map importance weight based on the noise simulation data, and uses the gated unit to filter the low-correlation feature channels; The signal output submodule extracts the effective signal area through adaptive threshold segmentation based on the weighted feature map and generates a corrected detection signal.

7. The AI ​​multi-task scheduling colloidal gold multi-index synchronous detection system according to claim 1 is characterized in that: The task scheduling engine module includes an evaluation matrix construction submodule, a resource allocation submodule, and an instruction generation submodule; The evaluation matrix construction submodule defines the dual-dimensional evaluation indicators of computational complexity and timeliness based on the detection task deadline constraint, and generates a task priority matrix; The resource allocation submodule uses the Hungarian algorithm to solve the optimal task-resource matching solution based on the task priority matrix and verifies the feasibility of the solution through time window constraints; The instruction generation submodule is based on a matching scheme and adopts an event-driven architecture to trigger a detection channel enable signal to generate a parallel detection instruction.

8. The AI ​​multi-task scheduling colloidal gold multi-index synchronous detection system according to claim 1 is characterized by: The cross-interference compensation module includes a tensor construction submodule, a pattern decomposition submodule, and a matrix generation submodule; The tensor construction submodule organizes the spectral dimension, time dimension and spatial dimension into a third-order tensor structure based on multiple batches of historical detection data; The pattern decomposition submodule is based on a third-order tensor structure, applies the CP decomposition algorithm to extract common interference basis vectors, and optimizes the decomposition residuals through the alternating least squares method; The matrix generation submodule reconstructs the current detection interference component based on the interference basis vector using sparse coding to generate a compensation coefficient matrix.

9. The AI ​​multi-task scheduling colloidal gold multi-index synchronous detection system according to claim 1 is characterized in that: The feedback module includes a sequence planning submodule, a parameter adjustment submodule, and an instruction generation submodule; The sequence planning submodule uses Petri nets to model the microvalve state transition logic based on the reaction chamber topology to generate the optimal action sequence; The parameter adjustment submodule uses a model predictive control algorithm to rolling optimize the residence time setting value based on real-time impedance phase angle data; The instruction generation submodule converts the voltage control amount through a pulse width modulation encoder based on the set value to generate a feedback control instruction.

10. The AI ​​multi-task scheduling colloidal gold multi-index synchronous detection system according to claim 4 is characterized in that: The nonlinear mapping submodule adopts an incremental training strategy, updates the network hidden layer nodes in real time through an online sequence extreme learning machine, and combines regularization parameters to prevent overfitting.

Images