Pesticide physicochemical property determination method

Abstract

The invention discloses a pesticide physicochemical property determination method, and relates to the technical field of pesticide physicochemical property determination, according to the method, 100 pL + / -10% volume of isotope labeling internal standard is injected into each sample in real time by using a piezoelectric driving microdroplet generator, automatic and continuous calibration labeling is realized, and a standard curve does not need to be repeatedly prepared; the Bayesian regression algorithm automatically generates and updates a calibration curve according to the real-time peak area ratio, and outputs the concentration prediction uncertainty at the same time, so that the traditional one-time calibration mode is replaced, and the quantitative precision and reliability are greatly improved; local calibration parameters and residual errors of all laboratories are encrypted and then uploaded to a central server, a global calibration model is formed through aggregation of a FedAvg federated average algorithm, and then the global calibration model is issued to all the laboratories for use; according to the method, data privacy is guaranteed, calibration models of different laboratories are highly consistent, rapid migration and version iteration of the calibration models are achieved, and the problem that calibration curves of multiple laboratories are inconsistent is thoroughly solved.

Description

Technical Field

[0001] The invention relates to the technical field of pesticide physicochemical property determination, in particular to a method for determining the physicochemical property of a pesticide. Background Art

[0002] As food safety supervision continues to increase, accurately understanding the physical and chemical properties of pesticides in the environment and agricultural products has become an important part of daily laboratory work; traditional measurement processes often start with the rapid sample QuEChERS pretreatment of agricultural products, and then complete multi-residue analysis through gas chromatography-mass spectrometry GC-MS / MS or liquid chromatography-mass spectrometry LC-MS / MS; this combination has been used in the simultaneous determination of multiple pesticides because of its efficiency and sensitivity, but in actual operation, the reliability of the test results depends to a large extent on the accuracy of the calibration curve, while the calibration curve itself is deeply affected by multiple factors such as matrix effect, pretreatment details, and instrument status.

[0003] A common practice is to use matrix-matched calibration, adding standard substances to the same matrix as the sample to establish a standard curve to offset background interference; however, there are slight differences in the sample sources, purification reagent package batches and even room temperature and humidity used by different laboratories, resulting in different peak area responses at the same concentration; although the internal standard method can partially compensate for extraction losses, it cannot correct the matrix response differences of all targets at the same time; although the standard addition method can reduce matrix interference, it increases the number of operation steps and time consumption.

[0004] When multiple laboratories test the same sample, they will find that although the operating procedures are consistent, there will be obvious deviations in the slope and intercept of the calibration curve. Such deviations not only make it difficult to directly compare the quantitative results between different laboratories, but also make it difficult for laboratories to conduct proficiency testing and establish shared databases. What's more, when there is a need to quickly expand to new matrices or new instrument platforms, the calibration model must be re-established almost every time, resulting in waste of resources and inefficiency.

[0005] In short, while the current methods for determining the physicochemical properties of pesticides meet the needs of high sensitivity and multi-residue analysis, they are stuck in the dilemma of calibration dependence and independence; the lack of portability of calibration curves not only increases the cost of internal quality control in laboratories, but also hinders the consistency and comparability of data across regions and platforms, bringing hidden dangers to regulatory decisions and risk assessments. How to build a calibration model that can be seamlessly transferred between different laboratories, different matrices, and different instruments while maintaining accuracy has become a core problem that needs to be solved urgently. Summary of the invention

[0006] In view of the above existing problems, the present invention is proposed.

[0007] The present invention provides a method for determining the physical and chemical properties of pesticides, which solves the problem that in the current methods for determining the physical and chemical properties of pesticides, while meeting the requirements of high sensitivity and multi-residue analysis, there is a lack of effective cross-regional interconnection in the calibration of different laboratories.

[0008] To solve the above technical problems, the present invention provides the following technical solutions:

[0009] An embodiment of the present invention provides a method for determining the physical and chemical properties of pesticides, which includes,

[0010] Step S1, taking agricultural products or environmental matrix samples, and performing pretreatment according to the quick sample pretreatment QuEChERS method for agricultural products to obtain a purified matrix extract;

[0011] Step S2, loading the matrix extract into an integrated microfluidic chip to obtain a mixed solution;

[0012] Step S3, continuously transporting the mixed solution generated in step S2 to a gas chromatography-tandem mass spectrometry GC-MS / MS detection system through a microfluidic interface, and synchronously collecting the peak area data of target pesticides and internal standards in the dynamic multiple reaction monitoring dMRM mode;

[0013] Step S4, according to the peak area ratio, automatically generating and updating a matrix-matched calibration curve by using the Bayesian regression algorithm;

[0014] Step S5, encrypting and uploading the local calibration curve parameters and detection residuals to a central server; the central server generates a global calibration model by fusing calibration parameters from different laboratories based on the federated learning framework;

[0015] Step S6, regularly issuing the global calibration model to each laboratory for replacing or optimizing the local Bayesian regression model to achieve the transferability of the calibration model across laboratories, instruments, and matrices.

[0016] As a preferred scheme of the method for determining the physical and chemical properties of pesticides according to the present invention, wherein: in step S2, an inkjet microdroplet generator is provided in the chip for injecting an isotope-labeled internal standard solution with a known mass concentration into the matrix extract at an accuracy of 100 pL ± 10% to add the internal standard in real time and obtain a mixed solution;

[0017] The microdroplet generator adopts a piezoelectric-driven micro ink droplet ejection unit, and its main structure consists of three parts: a nozzle array plate, a piezoelectric actuator, and a liquid supply module:

[0018] The nozzle array plate is a silicon-based substrate. Circular microholes with a single-hole aperture of 30 μm are prepared by SU-8 lithography. A microfluid chamber with a diameter of 150 μm and a length of 200 μm is connected to the rear end of each nozzle; the liquid channel is formed by soft lithography of polydimethylsiloxane (PDMS), with an overall height of 50 μm; the piezoelectric actuator selects a PZT ceramic sheet with a thickness of 150 μm and an area of 3×3 mm 2 and realizes single-shot injection through a 50 V peak value and 50 μs pulse drive. Each pulse can release ≈100 pL droplets, and the CV < 5%;

[0019] The driving method adopts a bipolar pulse controller, and the output pulse frequency is adjustable from 0 to 10 kHz; the working process of the nozzle is as follows: the pulse causes the piezoelectric sheet to generate an instantaneous depression, and the liquid is squeezed through the microhole to form droplets; then the piezoelectric sheet resets and generates a negative pressure to ensure that there is no residual liquid hanging on the wall; the outer surface of the nozzle is treated with a fluorosilane hydrophobic coating, and the static contact angle > 110° to prevent liquid backflow;

[0020] Volume calibration is carried out jointly by the gravimetric method and the high-speed imaging method: First, in an environment with a constant temperature of 23 ± 0.5 °C and a constant humidity of 50 ± 2% RH, continuously inject 10,000 droplets onto a high-precision balance of ±0.01 μg, calculate the mass of a single droplet and convert it into volume; Second, use a 10,000 fps high-speed camera to record the injection process and compare the optical cross-sectional area to confirm the volume accuracy; Only when the deviation between the measured values of the two methods < 3% can the calibration be confirmed to be completed;

[0021] The connection with the chip flow channel adopts a plug-and-play liquid interface: the nozzle array plate is tightly attached to the Cyclic Olefin Copolymer (COC) substrate through a threaded fixing seat, and the O-ring fluororubber seal can withstand a pressure ≤ 15 psi to prevent leakage; the nozzle orifice is aligned to the bottom of the chip main flow channel, and the flow channel cross-sectional area is 300×50 μm 2 to form an integrated fluid network with the nozzle microfluid chamber; the chip can be replaced without disassembling the nozzle module; at the same time, it supports the rapid access of a standard 360 μm OD PEEK hose to the sample input port.

[0022] As a preferred solution of the method for determining the physical and chemical properties of pesticides described in the present invention, wherein: in step S3, the mixed solution is continuously transported to the gas chromatography tandem mass spectrometry GC-MS / MS detection system through the microfluidic interface, and the peak area data of the target pesticide and the internal standard are collected in real time. Specifically:

[0023] A continuous coupling transport channel is established at the microfluidic interface, and the mixed solution is injected into the GC spray injection port at a constant volumetric flow rate, so as to realize the seamless docking of the sample and the detection system. The continuous injection volume is expressed as:

[0024] V = Q × t inj ,

[0025] Among them, V represents the continuous injection volume, Q represents the output volume flow rate of the microfluidic interface, and t inj represents the effective time of continuous injection of GC-MS / MS. The effective volume entering the chromatographic column is regulated by the split ratio and is expressed as:

[0026]

[0027] Among them, V eff represents the effective injection volume entering the chromatographic column, and S represents the split ratio of the GC interface.

[0028] As a preferred scheme of the method for determining the physical and chemical properties of pesticides described in the present invention, in step S3, during the detection stage, the MS / MS system monitors the parent-child ion conversion reaction of each target in the dynamic multiple reaction monitoring dMRM mode, and the total single-cycle monitoring time is expressed as:

[0029]

[0030] Among them, t cycle represents the total single-cycle monitoring time of MS / MS, n represents the total number of dynamic monitoring reactions, d j represents the residence time of the jth monitoring reaction, and t delay represents the reaction conversion delay time.

[0031] As a preferred scheme of the method for determining the physical and chemical properties of pesticides described in the present invention, in step S3, the calculation formula for the peak area of each substance is: A i = k i c i V eff η i , where A i represents the peak area of the ith substance, k i represents the instrument response factor, c i represents the concentration of the ith substance in the mixed solution, and η i represents the transmission coupling efficiency, reflecting the overall efficiency of substance transmission from the microfluidic interface to the detection system;

[0032] The calculation formula for the peak area ratio of the target pesticide to the internal standard obtained by real-time acquisition is: R = A1 / A2, where R represents the peak area ratio of the target pesticide to the internal standard, A1 represents the peak area of the target pesticide, and A2 represents the peak area of the internal standard.

[0033] As a preferred scheme of the method for determining the physical and chemical properties of pesticides described in the present invention, in step S4, the matrix-matched calibration curve includes a linear fitting coefficient and an uncertainty evaluation parameter.

[0034] As a preferred embodiment of the method for determining the physical and chemical properties of a pesticide according to the present invention, in step S4, a calibration curve based on matrix matching is constructed by the Bayesian regression algorithm;

[0035] The relationship between the peak area ratio and the concentration of the target pesticide is described by a linear model, and the mathematical model is expressed as:

[0036] R = β0 + β1c + ∈,

[0037] where R represents the peak area ratio, β0 represents the intercept, β1 represents the slope, c represents the concentration of the target pesticide in the calibration sample, and ∈ represents an error term that follows a normal distribution;

[0038] A prior distribution is selected to constrain the parameters, and it is assumed that the parameter vector follows a multivariate normal distribution:

[0039]

[0040] where, represents the linear model parameters, μ0 represents the prior mean vector, and Σ0 represents the prior covariance matrix.

[0041] As a preferred embodiment of the method for determining the physical and chemical properties of a pesticide according to the present invention, in step S4, the step of constructing a calibration curve based on matrix matching by the Bayesian regression algorithm further includes:

[0042] Under the condition of the observed data each row of the design matrix X is constructed as x i =(1, c i ), and let Based on the conjugacy of the likelihood function and the prior distribution, the posterior update formula is:

[0043] where Σ n represents the posterior covariance matrix, μ n represents the posterior mean vector, σ 2 represents the variance of the error term, X represents the design matrix, X T represents the transpose of X, and R represents the observed peak area ratio vector;

[0044] To evaluate the uncertainty of the calibration curve, variance prediction is performed, and the prediction variance formula is:

[0045]

[0046] where, represents the prediction variance at the new concentration c, represents the eigenvector corresponding to the new concentration, and Σ n represents the posterior covariance matrix;

[0047] σ 2 represents the observation noise variance.

[0048] As a preferred solution of the method for determining the physical and chemical properties of a pesticide according to the present invention, wherein: in step S5, each laboratory first encrypts the local calibration curve parameters and the detection residuals, and then uploads them to the central server. The steps include:

[0049] Each laboratory records the local calibration curve parameters as θ i and the detection residuals as ε i , θ i includes the local regression coefficient vector obtained from step S4, and forms encrypted data after being processed by an encryption function. The mathematical expression is:

[0050] E l = ε(θ l , ε l ; k e ),

[0051] wherein, E l represents the encrypted data uploaded by laboratory l, ε represents the encryption function, θ l represents the local calibration curve parameters, ε l represents the detection residuals, and k e represents the key used for encryption.

[0052] As a preferred solution of the method for determining the physical and chemical properties of a pesticide according to the present invention, wherein: in step S5, the central server receives the encrypted data of each laboratory through a secure transmission channel, and completes the decryption process in a trusted execution environment to obtain the original local parameters. Subsequently, the central server uses the federated averaging algorithm FedAvg to aggregate the data of each laboratory to generate a global calibration model. The aggregation formula is:

[0053]

[0054] wherein, θ g represents the global calibration model parameters, w i represents the weight of laboratory i, n i represents the number of samples of laboratory i, and N represents the total number of laboratories participating in federated learning;

[0055] The global estimation of the detection residuals is calculated by weighted average, and the expression is:

[0056] wherein, ε g represents the global detection residuals;

[0057] After generating the global model, the central server distributes the global calibration model to each laboratory through a secure encrypted transmission channel.

[0058] The beneficial effects of the present invention are as follows:

[0059] By combining microfluidic inkjet microdrop internal standard addition, Bayesian dynamic calibration, and federated learning, the present invention constructs a pesticide physical and chemical property determination system that can be seamlessly migrated across laboratories, instruments, and matrices, effectively overcoming the problem that traditional calibration curves cannot be universal due to differences in experimental conditions.

[0060] The present invention uses a piezoelectric-driven microdrop generator to inject 100 pL ± 10% volume of isotope-labeled internal standard into each sample in real time, realizing automated and continuous calibration labeling without the need to repeatedly prepare standard curves; the Bayesian regression algorithm automatically generates and updates the calibration curve based on the real-time peak area ratio, and at the same time outputs the concentration prediction uncertainty, replacing the traditional one-time calibration method, greatly improving the quantitative accuracy and reliability.

[0061] The present invention uploads the local calibration parameters and residuals of each laboratory to the central server after encryption, aggregates them through the FedAvg federated averaging algorithm to form a global calibration model, and then distributes it to each laboratory for use; it not only ensures data privacy, but also makes the calibration models between different laboratories highly consistent, realizing the rapid migration and version iteration of the calibration model, and completely solving the problem of inconsistent calibration curves in multiple laboratories.

[0062] The present invention significantly saves experimental consumables and labor costs: microdrop internal standard addition reduces the QuEChERS extraction volume to the microliter level, reducing the usage of solvents and standards; the shortened continuous injection design reduces the single-sample detection time, suitable for high-throughput applications; Bayesian uncertainty assessment can quantify the credible interval of the results, replacing the rough assessment method that only reports recovery rates and RSDs.

[0063] The present invention has strong compatibility and can quickly adapt to new pesticides, new matrices, and new instruments. Only a small amount of local samples are needed for verification to deploy the global calibration model, significantly improving the method development efficiency; overall, the present invention constructs an efficient, accurate, migratable, and secure pesticide physical and chemical property determination platform, providing a unified and reliable data basis for regulatory agencies and testing laboratories. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0065] Figure 1 It is a schematic flow chart of the pesticide physical and chemical property determination method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following specifically describes the embodiments of the present invention in detail with reference to the accompanying drawings of the specification.

[0067] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0068] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not all refer to the same embodiment, nor is it an embodiment that is separate from or mutually exclusive of other embodiments.

[0069] Example 1, referring to Figure 1 , this example provides a method for measuring the physical and chemical properties of pesticides, including the following steps:

[0070] Step S1: Take samples of agricultural products or environmental matrices, and perform pretreatment according to the quick sample pretreatment QuEChERS method for agricultural products to obtain a purified matrix extract.

[0071] Step S2: Load the matrix extract onto an integrated microfluidic chip to obtain a mixed solution.

[0072] In step S2, an inkjet microdroplet generator is provided inside the chip, which is used to inject an isotope-labeled internal standard solution with a known mass concentration into the matrix extract with an accuracy of 100 pL ± 10% to add the internal standard in real time and obtain a mixed solution.

[0073] The microdroplet generator adopts a piezoelectric-driven micro ink droplet ejection unit, and its main structure consists of three parts: a nozzle array plate, a piezoelectric actuator, and a liquid supply module.

[0074] The nozzle array plate is a silicon-based substrate, and circular micropores with a single-hole aperture of 30 μm are prepared by SU-8 lithography. A microfluid chamber with a diameter of 150 μm and a length of 200 μm is connected to the rear end of each nozzle; the liquid channel is formed by soft etching of polydimethylsiloxane (PDMS), and the overall height is 50 μm; the piezoelectric actuator selects a PZT ceramic sheet with a thickness of 150 μm and an area of 3 × 3 mm 2 , and single-shot ejection is achieved by driving with a 50 V peak value and a 50 μs pulse. Each pulse can release ≈100 pL droplets, and CV < 5%.

[0075] The driving method adopts a bipolar pulse controller, and the output pulse frequency is adjustable from 0 to 10 kHz; the working process of the nozzle is as follows: the pulse causes the piezoelectric sheet to produce an instantaneous depression, and the liquid is squeezed through the micropores to form droplets; then the piezoelectric sheet resets and generates negative pressure to ensure that there is no residual liquid hanging on the wall; the outer surface of the nozzle is treated with a fluorosilane hydrophobic coating, and the static contact angle > 110° to prevent liquid backflow.

[0076] Volume calibration is carried out jointly by the gravimetric method and the high-speed imaging method: First, in an environment with a constant temperature of 23 ± 0.5 °C and a constant humidity of 50 ± 2% RH, continuously spray 10,000 drops onto a high-precision balance of ±0.01 μg, calculate the mass of a single drop and convert it into volume; secondly, use a 10,000 fps high-speed camera to record the spraying process and compare the optical cross-sectional area to confirm the volume accuracy; only when the deviation between the measurement values of the two methods < 3% can the calibration be confirmed to be completed;

[0077] The connection with the chip flow channel adopts a plug-and-play liquid interface: the nozzle array plate is tightly attached to the Cyclic Olefin Copolymer (COC) substrate through a threaded fixing seat, and the O-ring fluororubber seal can withstand a pressure ≤ 15 psi to prevent leakage; the nozzle orifice is aligned to the bottom of the chip main flow channel, and the flow channel cross-sectional area is 300 × 50 μm 2 , forming an integrated fluid network with the nozzle microfluidic chamber; the chip can be replaced without disassembling the nozzle module; at the same time, it supports the rapid access of a standard 360 μm OD PEEK hose to the sample input port.

[0078] Step S3, continuously transport the mixed solution generated in step S2 to the gas chromatography-tandem mass spectrometry GC-MS / MS detection system through the microfluidic interface, and synchronously collect the peak area data of the target pesticide and the internal standard in the dynamic multiple reaction monitoring dMRM mode;

[0079] In step S3, the mixed solution is continuously transported to the gas chromatography tandem mass spectrometry GC-MS / MS detection system through the microfluidic interface, and the peak area data of the target pesticide and the internal standard are collected in real time. Specifically:

[0080] Establish a continuous coupling transport channel at the microfluidic interface, and inject the mixed solution into the GC spray injection port at a constant volume flow rate, so as to achieve seamless docking between the sample and the detection system. The continuous injection volume is expressed as:

[0081] V = Q × t inj ,

[0082] Among them, V represents the continuous injection volume, Q represents the output volume flow rate of the microfluidic interface, and t inj represents the effective time of continuous injection of GC-MS / MS. The effective volume entering the chromatographic column is regulated by the split ratio and is expressed as:

[0083]

[0084] Among them, V eff represents the effective injection volume entering the chromatographic column, and S represents the split ratio of the GC interface;

[0085] In step S3, during the detection stage, the MS / MS system monitors the parent-daughter ion conversion reaction of each target in the dynamic multiple reaction monitoring (dMRM) mode. The total time of a single cycle of monitoring is expressed as:

[0086]

[0087] Among them, t cycle represents the total time of a single cycle of MS / MS monitoring, n represents the total number of dynamic monitoring reactions, d j represents the residence time of the j-th monitoring reaction, and t delay represents the reaction conversion delay time;

[0088] In step S3, the calculation formula for the peak area of each substance is: A i = k i c i V eff η i Among them, A i represents the peak area of the i-th substance, k i represents the instrument response factor, c i represents the concentration of the i-th substance in the mixed solution, and η i represents the transmission coupling efficiency, reflecting the overall efficiency of substance transmission from the microfluidic interface to the detection system;

[0089] The calculation formula for the peak area ratio of the target pesticide to the internal standard obtained by real-time acquisition is: R = A1 / A2. Among them, R represents the peak area ratio of the target pesticide to the internal standard, A1 represents the peak area of the target pesticide, and A2 represents the peak area of the internal standard;

[0090] Specifically, the process of continuously transporting the sample to the GC-MS / MS system and real-time collecting the peak area data of the target pesticide and the internal standard in the dynamic multiple reaction monitoring mode is shown here; by establishing a mathematical model of volume injection, splitting, and dynamic monitoring, it is possible to ensure the accuracy of the sample entering the chromatographic column and the synchronization of MS / MS detection under continuous injection conditions;

[0091] Step S4, according to the peak area ratio, automatically generate and update the matrix-matched calibration curve using the Bayesian regression algorithm;

[0092] In step S4, the matrix-matched calibration curve includes linear fitting coefficients and uncertainty evaluation parameters;

[0093] In step S4, construct the matrix-matched calibration curve through the Bayesian regression algorithm;

[0094] The relationship between the peak area ratio and the concentration of the target pesticide is described by a linear model, and the mathematical model is expressed as:

[0095] R = β0 + β1c + ∈,

[0096] where R represents the peak area ratio, β0 represents the intercept, β1 represents the slope, c represents the concentration of the target pesticide in the calibration sample, and ∈ represents an error term that follows a normal distribution;

[0097] A prior distribution is selected to constrain the parameters, and it is assumed that the parameter vector follows a multivariate normal distribution:

[0098]

[0099] where, represents the linear model parameters, μ0 represents the prior mean vector, and Σ0 represents the prior covariance matrix;

[0100] In step S4, the steps of constructing a matrix-matched calibration curve by the Bayesian regression algorithm further include:

[0101] Under the condition of the observed data , each row of the design matrix X is constructed as x i = (1, c i ), and let Based on the conjugacy of the likelihood function and the prior distribution, the posterior update formula is:

[0102] where Σ n represents the posterior covariance matrix, μ n represents the posterior mean vector, σ 2 represents the variance of the error term, X represents the design matrix, X T represents the transpose of X, and R represents the observed peak area ratio vector;

[0103] To evaluate the uncertainty of the calibration curve, variance prediction is performed, and the prediction variance formula is:

[0104]

[0105] where, represents the prediction variance at the new concentration c, represents the eigenvector corresponding to the new concentration, and Σ n represents the posterior covariance matrix;

[0106] σ 2 represents the observed noise variance;

[0107] Specifically, a calibration curve is established here through the Bayesian regression method. The prior information is used to provide initial constraints for the model parameters, and the posterior distribution of the parameters is continuously updated through the observed data to achieve automatic model update;

[0108] The linear model characterizes the quantitative relationship between the peak area ratio and the concentration. The prior distribution ensures the rationality of the parameters when the data is insufficient. The posterior update integrates the prior knowledge and the observed evidence, so as to obtain accurate calibration parameters. The calculation of the prediction variance not only reflects the adaptability of the model to new data, but also provides a quantitative basis for uncertainty evaluation;

[0109] Step S5: Encrypt and upload the local calibration curve parameters and the detection residuals to the central server; The central server generates a global calibration model by fusing the calibration parameters from different laboratories based on the federated learning framework;

[0110] In step S5, each laboratory first encrypts the local calibration curve parameters and the detection residuals, and then uploads them to the central server. The steps include:

[0111] Each laboratory records the local calibration curve parameters as θ i and the detection residuals as ε i , θ i including the local regression coefficient vector obtained from step S4, which forms encrypted data after being processed by the encryption function. The mathematical expression is:

[0112] E l = ε(θ l , ε l ; k e ),

[0113] where, E l represents the encrypted data uploaded by laboratory l, ε represents the encryption function, θ l represents the local calibration curve parameters, ε l represents the detection residuals, and k e represents the encryption key;

[0114] In step S5, the central server receives the encrypted data from each laboratory through a secure transmission channel, and completes the decryption process in a trusted execution environment to obtain the original local parameters. Subsequently, the central server uses the federated average algorithm FedAvg to aggregate the data of each laboratory to generate a global calibration model. The aggregation formula is:

[0115]

[0116] where, θ g represents the global calibration model parameters, w i represents the weight of laboratory i, and n i$n_i$ represents the number of samples in laboratory $i$, and $N$ represents the total number of laboratories participating in federated learning;

[0117] The global estimate of the detection residual is calculated by weighted average and expressed as:

[0118] where $\epsilon$ g represents the global detection residual;

[0119] After generating the global model, the central server distributes the global calibration model to each laboratory through a secure encrypted transmission channel;

[0120] Specifically, each laboratory protects the data privacy of the calibration parameters and detection residuals through local encryption technology, and then uploads the encrypted data to the central server using a secure transmission protocol. After the central server decrypts the data securely, it synthesizes the data of each laboratory through the federated averaging algorithm to form a global calibration model;

[0121] This aggregation process uses the number of samples in each laboratory as weights to achieve reasonable data fusion, and at the same time, the detection residuals are also globally evaluated using weighted average;

[0122] Step S6, the global calibration model is regularly distributed to each laboratory to replace or optimize the local Bayesian regression model, so as to achieve the transferability of the calibration model across laboratories, instruments, and matrices;

[0123] The core of the present invention lies in using a droplet generator to complete the automatic update of the calibration curve and the fusion of the federated model after real-time internal standard addition.

[0124] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limitations. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for determining the physicochemical properties of pesticides, characterized in that: include, Step S1, taking agricultural product or environmental matrix samples, and pre-treating them according to the rapid sample pre-treatment QuEChERS method for agricultural products to obtain a purified matrix extract; Step S2, loading the matrix extract into an integrated microfluidic chip to obtain a mixed solution; Step S3, continuously delivering the mixed solution generated in step S2 to a gas chromatography-tandem mass spectrometry (GC-MS / MS) detection system through a microfluidic interface, and synchronously collecting target pesticide and internal standard peak area data in a dynamic multiple reaction monitoring (dMRM) mode; Step S4, automatically generating and updating a calibration curve based on matrix matching using a Bayesian regression algorithm according to the peak area ratio; Step S5, uploading the local calibration curve parameters and the detection residuals to the central server after encryption; the central server fuses the calibration parameters from different laboratories based on the federated learning framework to generate a global calibration model; Step S6: The global calibration model is regularly sent to each laboratory to replace or optimize the local Bayesian regression model.

2. A method for determining the physicochemical properties of pesticides as claimed in claim 1, characterized in that: In step S2, an inkjet droplet generator is provided in the chip, which is used to inject an isotope-labeled internal standard solution of known mass concentration into the matrix extract, and add the internal standard in real time to obtain a mixed solution.

3. A method for determining the physicochemical properties of pesticides as claimed in claim 1, characterized in that: In step S3, the mixed solution is continuously transported to the gas chromatography tandem mass spectrometry GC-MS / MS detection system through the microfluidic interface to collect the peak area data of the target pesticide and the internal standard in real time. Specifically: A continuous coupling delivery channel was established on the microfluidic interface, and the mixed solution was injected into the GC spray injection port at a constant volume flow rate. The continuous injection volume was expressed as: V=Q×t inj , Where V represents the continuous injection volume, Q represents the output volume flow rate of the microfluidic interface, and t inj It represents the effective time of GC-MS / MS continuous injection. The effective volume entering the chromatographic column is regulated by the split ratio and is expressed as: Among them, V eff It represents the effective injection volume into the chromatographic column, and S represents the split ratio of the GC interface.

4. A method for determining the physicochemical properties of pesticides as claimed in claim 3, characterized in that: In step S3, during the detection phase, the MS / MS system performs parent-daughter ion conversion reaction monitoring on each target in the dynamic multiple reaction monitoring dMRM mode, and the total monitoring time of a single cycle is expressed as: Among them, t cycle represents the total time of single cycle monitoring of MS / MS, n represents the total number of dynamic monitoring reactions, d j represents the residence time of the jth monitored reaction, t delay Indicates the reaction conversion delay time.

5. A method for determining the physicochemical properties of pesticides as claimed in claim 4, characterized in that: In step S3, the peak area calculation formula for each substance is: i =k i c i V eff η i , where A i represents the peak area of ​​the i-th substance, k i represents the instrument response factor, c i represents the concentration of the i-th substance in the mixed solution, η i It represents the transmission coupling efficiency, which reflects the overall efficiency of material transfer from the microfluidic interface to the detection system; The calculation formula for the peak area ratio of the target pesticide and the internal standard obtained by real-time collection is: R=A1 / A2, where R represents the peak area ratio of the target pesticide and the internal standard, A1 represents the peak area of ​​the target pesticide, and A2 represents the peak area of ​​the internal standard.

6. A method for determining the physicochemical properties of pesticides as claimed in claim 1, characterized in that: In step S4, the calibration curve based on matrix matching includes linear fitting coefficients and uncertainty evaluation parameters.

7. A method for determining the physicochemical properties of pesticides as claimed in claim 6, characterized in that: In step S4, a calibration curve based on matrix matching is constructed by a Bayesian regression algorithm; A linear model was used to describe the relationship between the peak area ratio and the target pesticide concentration. The mathematical model was expressed as: R=β0+β1c+∈, Where R represents the peak area ratio, β0 represents the intercept, β1 represents the slope, c represents the concentration of the target pesticide in the calibration sample, and ∈ represents the error term that follows the normal distribution; Select a priori distribution to constrain the parameters, assuming that the parameter vector obeys a multivariate normal distribution: in, represents the linear model parameters, μ0 represents the prior mean vector, and Σ0 represents the prior covariance matrix.

8. A method for determining the physicochemical properties of pesticides as claimed in claim 7, characterized in that: In step S4, the step of constructing a calibration curve based on matrix matching by Bayesian regression algorithm also includes: In the observation data Under the condition of i =(1,c i ), and order Based on the conjugation of the likelihood function and the prior distribution, the posterior update formula is: Among them, Σ n represents the posterior covariance matrix, μ n represents the posterior mean vector, σ 2 represents the variance of the error term, X represents the design matrix, and X T represents the transpose of X, and R represents the observed peak area ratio vector; In order to evaluate the uncertainty of the calibration curve, the variance prediction is performed, and the predicted variance formula is: in, represents the prediction variance at the new concentration c, represents the eigenvector corresponding to the new concentration, Σ n represents the posterior covariance matrix; σ 2 represents the observation noise variance.

9. A method for determining the physicochemical properties of pesticides as claimed in claim 1, characterized in that: In step S5, each laboratory first encrypts the local calibration curve parameters and detection residuals, and then uploads them to the central server. The steps include: Each laboratory records the local calibration curve parameter as θ i and the detection residual is denoted as ε i ,θ i The local regression coefficient vector obtained in step S4 is processed by the encryption function to form encrypted data, which is mathematically expressed as: E l =ε(θ l ,he l ;k e ), Among them, E l represents the encrypted data uploaded by laboratory l, ε represents the encryption function, θ l represents the local calibration curve parameter, ε l represents the detection residual, k e Represents the key used for encryption.

10. A method for determining the physicochemical properties of pesticides as claimed in claim 9, characterized in that: In step S5, the central server receives the encrypted data from each laboratory through a secure transmission channel, completes the decryption process in a trusted execution environment, and obtains the original local parameters. Then, the central server aggregates the data from each laboratory using the federated averaging algorithm FedAvg to generate a global calibration model. The aggregation formula is: Among them, θ g represents the global calibration model parameters, w i represents the weight of laboratory i, n i represents the number of samples of laboratory i, and N represents the total number of laboratories participating in federated learning; The global estimation of the detection residual is calculated by weighted average, which is expressed as: Among them, ε g represents the global detection residual; After the global model is generated, the central server sends the global calibration model to each laboratory through a secure encrypted transmission channel.

Images