CN-122017059-A - Optimization method for detection conditions of multi-target auxiliary agent in complex matrix

CN122017059ACN 122017059 ACN122017059 ACN 122017059ACN-122017059-A

Abstract

The invention relates to the technical field of analytical chemistry, in particular to an optimization method of multi-target auxiliary agent detection conditions in a complex matrix, which comprises the steps of initializing the detection conditions of multi-target auxiliary agent detection in the complex matrix, analyzing a complex matrix sample, collecting chromatographic/mass spectrum data, and extracting characteristic parameters of the chromatographic/mass spectrum data; the method comprises the steps of constructing a quantitative relation model between key modeling parameters and each performance index Y, training a depth Q network through an active learning algorithm based on the quantitative relation model and a multi-objective rewarding function, obtaining optimized key modeling parameters through the trained depth Q network, feeding the optimized key modeling parameters back to a detection system, and adjusting detection conditions of a gas chromatograph or a gas chromatograph-mass spectrometer in real time. The invention can realize real-time collaborative optimization and self-adaptive optimization of multiple detection conditions, and improves the detection efficiency, accuracy and robustness of multi-target auxiliary agent detection in complex matrixes.

Inventors

LI ZHIWEI
ZHANG LANLAN
Huang chuangfeng
CAI HAIMING
LAI CHAOKUN
Ding Enen
CHEN LIYI
XIANG YONG
ZHOU ZIGANG

Assignees

广电计量检测集团股份有限公司
广电计量检测（湖南）有限公司
广电计量检测（重庆）有限公司

Dates

Publication Date: 20260512
Application Date: 20251225

Claims (8)

1. The optimizing method of the detection condition of the multi-target auxiliary agent in the complex matrix is characterized by comprising the following steps: S1, initializing detection conditions for multi-target auxiliary agent detection in a complex matrix, analyzing a complex matrix sample, collecting chromatographic data and/or mass spectrum data, extracting characteristic parameters of the chromatographic data and/or mass spectrum data, and constructing a characteristic parameter data set based on the extracted characteristic parameters; s2, defining key modeling parameters and optimization targets, constructing a quantitative relation model between the key modeling parameters and each performance index Y, training a depth Q network through an active learning algorithm based on a characteristic parameter data set based on the quantitative relation model and a multi-target rewarding function, and outputting optimized key modeling parameters through the trained depth Q network; and S3, feeding the optimized key modeling parameters back to a detection system, and adjusting detection conditions of the gas chromatograph or the gas chromatograph-mass spectrometer in real time.
2. The method for optimizing detection conditions of multiple target aids in a complex matrix according to claim 1, wherein the step S1 comprises: S11, initializing detection conditions of a gas chromatograph or a gas chromatograph-mass spectrometer based on a history database of multi-target auxiliary agent detection in a complex matrix before analyzing the complex matrix sample; S12, analyzing the complex matrix sample by using a gas chromatograph or a gas chromatograph-mass spectrometer and collecting chromatographic or mass spectrum data in real time, extracting the following characteristic parameters, namely retention time, peak area, peak height, signal to noise ratio, separation degree, tailing factor and scanning range, and constructing a characteristic parameter data set based on the extracted characteristic parameters.
3. The method for optimizing detection conditions of multiple target aids in a complex matrix according to claim 1, wherein the step S2 comprises: s21, defining a state space and an action space of a depth Q network, wherein the state space comprises key modeling parameter combinations, and the action space is defined as an adjustment operation of detection conditions; S22, setting an optimization target, wherein the optimization target comprises higher separation degree, shorter analysis time and lower detection limit; s23, respectively constructing a response surface model for each performance index based on an optimization target, training each response surface model, and obtaining a trained response surface model, wherein the trained response surface model is used for outputting the corresponding performance index value according to the input key modeling parameters; S24, setting a multi-objective rewarding function of the depth Q network based on an optimization target, and solving by adopting an improved non-dominant sorting genetic algorithm to generate a Pareto optimal solution set; s25, training the depth Q network through an active learning algorithm based on a characteristic parameter data set and a multi-objective rewarding function by taking the trained response surface model as a reinforcement learning virtual environment; S26, defining a design space threshold, setting up standard thresholds for model fitting goodness and each performance index, and calculating the probability that key modeling parameters of each solution meet all standard thresholds based on the generated Pareto optimal solution set, so as to preferentially reserve high-probability solutions and obtain optimized key modeling parameters.
4. The method for optimizing multi-target auxiliary agent detection conditions in a complex matrix according to claim 3, wherein the key modeling parameters comprise a heating rate, a sample inlet temperature, a carrier gas flow rate, a column box initial temperature and an ion source temperature, and the action space is defined as an adjustment operation of the detection conditions and comprises discrete actions of adjusting the heating rate, the sample inlet temperature and the carrier gas flow rate.
5. The method for optimizing detection conditions of multiple target auxiliaries in a complex matrix according to claim 3, wherein the network structure of the deep Q network comprises an input layer, a hidden layer and an output layer, the input layer is used for receiving and outputting characteristic parameter vectors, the dimension is set according to the characteristic quantity, the hidden layer comprises 3 full-connection layers, 128 nodes of the first hidden layer, 64 nodes of the second hidden layer and 32 nodes of the third hidden layer, a ReLU activation function is used, the output layer is an action cost function Q (s, a), and the output dimension is equal to the action space size.
6. A method for optimizing detection conditions for multiple target adjuvants in a complex matrix according to claim 3, wherein said step S23 comprises: generating an experiment matrix by adopting a deterministic screening design method, covering all key modeling parameters of a parameter space with the least experiment times, recording the key modeling parameters and response values of each experiment, and generating a training sample set; and respectively constructing a response surface model for each performance index based on the training sample set, wherein the response surface model is used for describing the quantitative relation between the key modeling parameters and each performance index Y.
7. A method of optimizing multi-objective adjuvant detection conditions in a complex matrix according to claim 3, wherein said step of multi-objective rewarding function is expressed as: R=ω 1 * R _ R S +ω 2 * R_T+ω 3 * R_LOD; Wherein, R represents a comprehensive prize value, r_r S is a separation prize, r_t is an analysis time prize, r_lod represents a detection limit prize, x represents multiplication, and ω 1 、ω 2 、ω 3 are weight coefficients.
8. The method for optimizing multi-objective co-agent detection conditions in a complex matrix according to claim 6, wherein the training sample set comprises N samples, each sample comprising a key modeling parameter vector and a performance index vector.

Description

Optimization method for detection conditions of multi-target auxiliary agent in complex matrix Technical Field The invention relates to the technical field of analytical chemistry, in particular to an optimization method of detection conditions of multi-target auxiliary agents in a complex matrix. Background In the analytical chemistry field, detection of multi-target additives in complex matrices, such as environmental samples, biological fluids, industrial materials, including plasticizers, antioxidants, residual solvents, etc., is critical, and is widely used in food safety, environmental monitoring, and product quality control. Chromatographic/mass spectrometry techniques such as Gas Chromatography (GC) and gas chromatography-mass spectrometry (GC-MS) are the dominant detection means. However, a large amount of interfering components (such as proteins, lipids, inorganic salts, high molecular polymers, etc.) exist in the complex matrix, resulting in low separation efficiency of the target auxiliary agent and reduced detection sensitivity. The traditional detection method depends on empirical parameter setting, is difficult to adapt to dynamic changes of substrates with different samples, needs repeated manual trial and error optimization conditions, is time-consuming and labor-consuming, and has poor reproducibility. The prior art document CN119827682A discloses a method for detecting the folic acid metabolic substances of erythrocytes by using a liquid chromatography mass spectrometry tandem technique, which dynamically adjusts gradient parameters by monitoring the liquid chromatography in real time and utilizing a deep learning algorithm. However, the method is only aimed at specific biological samples (folic acid metabolites), does not involve detection of multi-target auxiliary agents in complex matrixes, is optimized based on image analysis, lacks systematic collaborative optimization of multi-detection conditions, does not introduce reinforcement learning or active learning mechanisms, does not realize adaptive learning based on real-time feedback, and does not involve dynamic optimization of detection conditions. With the development of artificial intelligence technology, machine learning has been applied to chromatographic data analysis, but the existing method focuses on offline modeling or single parameter adjustment, lacks real-time collaborative optimization of multiple detection conditions such as sample inlet temperature, temperature raising program, ion source temperature, scanning range and the like, and cannot realize dynamic closed-loop control of the detection process. Disclosure of Invention In order to solve the technical problems in the prior art, the invention provides an optimization method for multi-target auxiliary agent detection conditions in a complex matrix, which realizes real-time collaborative optimization and self-adaptive optimization of the multi-target auxiliary agent detection conditions by fusing reinforcement learning, active learning and multi-target optimization algorithm, and improves the detection efficiency, accuracy and robustness of multi-target auxiliary agent detection in the complex matrix. The invention aims to provide an optimization method for detection conditions of multiple target auxiliary agents in a complex matrix. The optimizing method of the detection condition of the multi-target auxiliary agent in the complex matrix comprises the following steps: s1, initializing detection conditions for multi-target auxiliary agent detection in a complex matrix, analyzing a complex matrix sample, collecting chromatographic/mass spectrum data, extracting characteristic parameters of the chromatographic/mass spectrum data, and constructing a characteristic parameter data set based on the extracted characteristic parameters; S2, defining key modeling parameters and optimization targets, constructing a quantitative relation model between the key modeling parameters and each performance index Y, training a depth Q network through an active learning algorithm based on a characteristic parameter data set based on the quantitative relation model and a multi-target rewarding function, and obtaining optimized key modeling parameters through the trained depth Q network; and S3, feeding the optimized key modeling parameters back to a detection system, and adjusting detection conditions of the gas chromatograph or the gas chromatograph-mass spectrometer in real time. Specifically, the step S1 includes: S11, initializing detection conditions of a gas chromatograph or a gas chromatograph-mass spectrometer based on a history database of multi-target auxiliary agent detection in a complex matrix before analyzing the complex matrix sample; S12, analyzing the complex matrix sample by using a gas chromatograph or a gas chromatograph-mass spectrometer and collecting chromatographic or mass spectrum data in real time, extracting the following characteristic parameters, namely rete