Prediction method and system of volatile organic compound chromatographic peak retention time

Abstract

The invention belongs to the field of organic compound monitoring, and particularly discloses a volatile organic compound chromatographic peak retention time prediction method and a volatile organic compound chromatographic peak retention time prediction system, and the method comprises the following steps: obtaining gas chromatographic data of an organic compound at a current moment and historical data of the organic compound; inputting the gas chromatography data and historical data into the prediction model to predict the retention time of the chromatographic peak of each substance of the organic compound at the current moment; the prediction model comprises a first model and a second model; the first model is used for acquiring a first predicted value of retention time based on the gas chromatography data; the second model is used for obtaining a second prediction value based on the historical data and the first prediction value, the second prediction value is the retention time predicted by the prediction model, or the second model is used for obtaining a second prediction value of the retention time based on the historical data; the retention time predicted by the prediction model is obtained by fusing the two predicted values. According to the method, the prediction precision of the retention time of the chromatographic peak of the organic compound is remarkably improved by fully utilizing multi-source information.

Description

Technical Field

[0001] The present application belongs to the field of organic compound monitoring, and more specifically, to a method and system for predicting the retention time of chromatographic peaks of volatile organic compounds. Background Art

[0002] Volatile organic compounds (VOCs) are important precursors of ozone and some secondary organic aerosols. Some of them are highly toxic to the environment and ecosystems. Therefore, VOCs have become one of the focuses of atmospheric environmental monitoring. Accurate and continuous long-term monitoring of VOCs is essential for ozone depletion, climate change, air quality monitoring, identification of air pollution sources, and public health early warning system research. Gas chromatography (GC) and gas chromatography-mass spectrometry (GC-MS) are commonly used instruments for monitoring atmospheric VOCs. The monitored substances are characterized in the form of chromatographic peaks in the gas chromatographic spectrum, where the retention time (RT) of the peak indicates the type of substance and the peak area reflects the concentration of the substance. Therefore, the positioning of the chromatographic peak (i.e., determining the retention time) is a key step in analyzing the gas chromatographic spectrum.

[0003] However, accurate manual or automated analysis of complex, continuous, long-term, industrial-scale VOCs monitoring data faces major challenges: (i) The complexity of environmental samples leads to the mixing of non-target and target substances, resulting in interfering peaks or overlapping peaks. (ii) Noise, baseline fluctuation or drift, and significant retention time drift may occur due to long-term continuous operation. (iii) In addition, the large differences in substance concentrations may result in extremely weak target peaks that are difficult to identify by visual inspection. The vast majority of these weak peaks are attributed to olefin compounds such as trans-2-butene, 1-butene, isobutylene, and cis-2-butene. These compounds have abnormally high ozone formation potentials (OFP), indicating that they make an important contribution to ozone formation. Inaccurate analysis of these key substances may have an adverse impact on the study of ozone formation mechanisms and the exploration of the causes of photochemical pollution. (iv) Due to the high complexity of chromatograms caused by the above factors, more than 6,000 monitoring stations in China mainly rely on manual analysis, which incurs huge human resource costs. In addition, manual analysis mainly relies on personal experience, which may lead to differences in analytical results between different sites due to personnel differences, thus affecting data consistency.

[0004] Therefore, the study of accurate and automated identification of chromatograms has attracted widespread attention. Existing studies have shown that machine learning, deep learning, or integrated models can be used to accurately predict the retention time of chromatographic peaks of monitored substances. However, the vast majority of studies treat each analysis as an isolated random event without considering the temporal connection between samples. In addition, although existing general chromatography analysis toolkits have demonstrated effective analytical capabilities in most laboratory analysis environments, these models receive information in some isolated form, such as only processing image or string information, or only analyzing information related to molecular structure. This shows that the model lacks the ability to use multimodal information for comprehensive analysis and cannot adapt to application scenarios that require diverse information input. Summary of the invention

[0005] In view of the defects of the prior art, the purpose of this application is to provide a method and system for predicting the retention time of chromatographic peaks of volatile organic compounds, aiming to solve the problems that the existing VOCs monitoring methods are single, fail to effectively integrate multiple information for monitoring, and have poor monitoring effects.

[0006] To achieve the above objectives, in a first aspect, the present application provides a method for predicting the retention time of a chromatographic peak of a volatile organic compound, comprising: Acquire the gas chromatography data of the organic compound at the current moment and the historical data of the organic compound; the historical data includes: the retention time data of each chromatographic peak of the substance in the gas chromatography measured in the past time period; The gas chromatography data and historical data are input into a prediction model to predict the retention time of each chromatographic peak of the organic compound at the current moment; the prediction model comprises: a first model and a second model; the first model is used to obtain a first prediction value of the retention time based on the gas chromatography data; the second model is used to obtain a second prediction value of the retention time based on the historical data and the first prediction value, and the second prediction value is the retention time predicted by the prediction model; or the second model is used to obtain a second prediction value of the retention time based on the historical data, and the retention time predicted by the prediction model is obtained by fusing the first prediction value and the second prediction value.

[0007] It should be noted that the present application comprehensively considers the gas chromatography data at the current moment and the historical data of retention time when predicting the retention time, fully refers to the current spectral characteristics and the historical migration characteristics of the material chromatographic peaks, and can effectively integrate the current spectral information and the historical migration characteristics of the retention time of the material chromatographic peaks, thereby achieving a more accurate retention time prediction; the above-mentioned scheme of the present application can predict the weak material chromatographic peaks that cannot be predicted by the existing retention time prediction relying on a single information. It can be seen that the retention time prediction method provided by the scheme of the present application has high accuracy and reliability.

[0008] In a possible implementation, the first model predicts a first predicted value of retention time based on gas chromatography data, including: The first model predicts the retention time of each substance chromatographic peak based on the gas chromatography data and the derivative data of the spectrum corresponding to the gas chromatography data to obtain a first prediction value; preferably, the derivative includes: a first-order derivative and a second-order derivative.

[0009] It is understood that in the present application, when the retention time is predicted based on gas chromatography, not only the chromatogram data of the chromatogram is considered, but also the derivatives at various locations in the chromatogram are calculated, and the retention time is predicted by combining the chromatogram data and the derivative (first-order derivative and second-order derivative) characteristics. Generally, the first-order derivative at the peak top of the chromatogram peak of a substance is 0, and the second-order derivative is less than 0.

[0010] In a possible implementation, the method further includes: Determining the first-order derivative of the retention time of the substance chromatographic peak predicted by the prediction model on the spectrum corresponding to the gas chromatography data; The retention time predicted by the prediction model is corrected according to the first-order derivative until the corresponding first-order derivative is 0, or until the corresponding first-order derivative is 0 and the second-order derivative is less than 0.

[0011] Furthermore, in theory, the predicted retention time should correspond to the top of the chromatographic peak of the substance, but there may be a certain deviation. Therefore, the first-order derivative and / or second-order derivative can be used to search, verify and adjust to ensure that the final retention time corresponds to the top of the chromatographic peak of the substance.

[0012] In a possible implementation, the first model uses a one-dimensional convolution kernel.

[0013] Among them, since the gas chromatography data is one-dimensional data, the first model uses a one-dimensional convolution kernel to adapt to the input data dimension to achieve retention time prediction based on gas chromatography data.

[0014] In one possible implementation, the prediction model is trained as a whole; or When the second model obtains the second prediction value of the retention time based on the historical data, the first model and the second model in the prediction model are trained separately, and then the prediction model is trained; when the second model obtains the second prediction value of the retention time based on the historical data and the first prediction value, the first model in the prediction model is trained separately, and then the prediction model is trained.

[0015] It should be noted that the prediction models obtained by using different training methods have been verified to have performance that can meet the expected use.

[0016] In a possible implementation, after the prediction model is trained using the training data of the first group of organic compounds: If the prediction model is used to predict the second group of organic compounds, when the second model obtains the second predicted value of the retention time based on the historical data, a small amount of training data of the second group of organic compounds is used to retrain the first model and the second model in the trained prediction model to update the prediction model, and then N dimensional adjustment modules are added to the updated prediction model so that the adjusted prediction model can be used to predict the second group of organic compounds; the number and / or type of material spectra included in the first group of organic compounds and the second group of organic compounds are different, N is an integer greater than or equal to 0, and when the number of the first group of organic compounds and the second group of organic compounds is the same, N is 0; If the prediction model is used to predict the second group of organic compounds, when the second model obtains the second prediction value of the retention time based on the historical data and the first prediction value, a small amount of training data of the second group of organic compounds is used to re-train the first model in the trained prediction model to update the prediction model, and then N dimensional adjustment modules are added to the updated prediction model so that the adjusted prediction model can be used to predict the second group of organic compounds; the number and / or type of material spectra included in the first group of organic compounds and the second group of organic compounds are different, N is an integer greater than or equal to 0, and when the number of the first group of organic compounds and the second group of organic compounds is the same, N is 0.

[0017] It can be understood that the prediction model provided in the present application takes into account the type and / or quantity of organic compound combinations at different sites and only makes a small adaptive migration of the prediction model, that is, it can effectively migrate it to the new site based on the original prediction model, thereby improving the migration performance of the prediction model and increasing the practicality of the prediction model.

[0018] In one possible implementation, the first model is a convolutional neural network; and / or the second model is a recurrent neural network or an attention mechanism type neural network.

[0019] In a second aspect, the present application provides a prediction system for chromatographic peak retention time of organic compounds, comprising: A data acquisition module, used to acquire gas chromatography data of organic compounds at the current moment and historical data of organic compounds; the historical data includes: retention time data of each chromatographic peak of a substance in the gas chromatography measured in the past time period; A retention time prediction module is used to input the gas chromatography data and historical data into a prediction model to predict the retention time of each chromatographic peak of the organic compound at the current moment; the prediction model includes: a first model and a second model; the first model is used to obtain a first prediction value of the retention time based on the gas chromatography data; the second model is used to obtain a second prediction value of the retention time based on the historical data and the first prediction value, and the second prediction value is the retention time predicted by the prediction model; or the second model is used to obtain a second prediction value of the retention time based on the historical data, and the retention time predicted by the prediction model is obtained by fusing the first prediction value and the second prediction value.

[0020] In one possible implementation, the first model used by the retention time prediction module predicts the retention time of each substance chromatographic peak based on the gas chromatography data and the derivative data of the spectrum corresponding to the gas chromatography data to obtain a first prediction value; preferably, the derivative includes: a first-order derivative and a second-order derivative.

[0021] In a third aspect, the present application provides an electronic device comprising: at least one memory for storing programs; and at least one processor for executing the programs stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.

[0022] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0023] In a fifth aspect, the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.

[0024] It can be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0025] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the prior art: The present application provides a method, system and device for predicting the retention time of volatile organic compounds. The proposed prediction model can simultaneously process real-time chromatograms and historical retention time series, make full use of multi-source information, and use the derivative uphill method to adjust the retention time predicted by multi-source information, which significantly improves the prediction accuracy. In a typical monitoring station experiment, the mean absolute error (MAE) of the prediction model provided by the present application in retention time positioning reached 0.0144 minutes, which is 2.76 to 38.19 times smaller than traditional machine learning or deep learning models.

[0026] The present application provides a method, system and device for predicting the retention time of volatile organic compounds, which can effectively adapt to various complex and abnormal chromatograms, including baseline fluctuations, baseline drifts and retention time drifts, and demonstrates strong robustness. This solves the challenges brought about by changes in the operating status of the instrument during long-term continuous monitoring.

[0027] The present application provides a method, system and device for predicting the retention time of volatile organic compounds, which makes full use of multi-source information and uses the derivative uphill method to adjust the retention time predicted by multi-source information, and can accurately identify weak chromatographic peaks that are usually difficult to distinguish visually, especially for olefin compounds with high ozone generation potential, such as trans-2-butene, 1-butene, isobutylene and cis-2-butene. This provides important support for in-depth research on the mechanism of ozone formation and helps to better understand and predict photochemical pollution.

[0028] The present application provides a method, system and device for predicting the retention time of volatile organic compounds, which has good migration ability and can adapt to the data characteristics of different monitoring sites through fine-tuning. In the cross-migration validation of multiple monitoring sites, the model showed strong migration performance. This means that the method has a wide range of industrial application potential and can be promoted and used in different regions and different types of monitoring sites.

[0029] The present application provides a method, system and device for predicting the retention time of volatile organic compounds, which realizes the automated high-precision analysis of VOCs monitoring data, greatly improves the monitoring efficiency and reduces the burden of manual analysis. This can not only save a lot of human resource costs, but also improve the consistency and reliability of data analysis, and provide more accurate data support for the study of VOCs pollution mechanism over a long time scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of a method for predicting chromatographic peak retention time of volatile organic compounds provided in an embodiment of the present application; Figure 2 is an example diagram of environmental monitoring data provided by an embodiment of the present application; FIG3( a ) is a schematic diagram of the architecture of the first prediction model provided in an embodiment of the present application; FIG3( b ) is a schematic diagram of the architecture of the second prediction model provided in an embodiment of the present application; FIG3( c ) is a schematic diagram of the architecture of the third prediction model provided in an embodiment of the present application; FIG3( d ) is a schematic diagram of the architecture of the fourth prediction model provided in an embodiment of the present application; Figure 4 It is a schematic diagram of the correction of retention time based on the derivative rise method provided in the embodiment of the present application; Figure 5 The following are examples of the retention time values ​​of real environment samples predicted by the embodiments of the present application: (a) samples of weak peaks and overlapping peaks at site D; (b) samples of baseline drift, weak peaks and overlapping peaks at site C; (c) samples of baseline jitter and weak peaks at site W; (d) samples of peak drift, weak peaks and overlapping peaks at site D; Figure 6 It is a schematic diagram of the prediction model migration provided by the embodiment of the present application; Figure 7 is an architecture diagram of a volatile organic compound chromatographic peak retention time prediction system provided in an embodiment of the present application; Figure 8 It is an architectural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0032] The term "and / or" in this article is a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The symbol " / " in this article indicates that the associated objects are in an or relationship, for example, A / B means A or B.

[0033] The terms "first", "second", etc. in the specification and claims herein are used to distinguish different objects rather than to describe a specific order of objects. For example, a first model and a second model are used to distinguish different models rather than to describe a specific order of models.

[0034] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0035] First, the technical terms involved in the embodiments of the present application are introduced.

[0036] (1) Retention time Retention time refers to the time from the start of sample injection to the appearance of the chromatographic peak. In gas chromatography analysis, retention time is an important basis for identifying substances. Each substance has its characteristic retention time under specific chromatographic conditions.

[0037] (2) First model In the embodiment of the present application, the first model is named more vividly according to its function: image processing layer (IPLs). In the embodiment of the present application, the image processing layer refers to a neural network layer for processing and analyzing gas chromatograms. In the present application, IPLs are mainly used to extract characteristic information of chromatograms.

[0038] (3) Second model In the embodiment of the present application, the first model is named more vividly according to its function: sequence processing layer (SPLs), which refers to a neural network layer used to process and analyze historical retention time series. In the present application, SPLs are mainly used to capture the temporal dynamic characteristics of retention time.

[0039] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.

[0040] Figure 1 is a flow chart of a method for predicting chromatographic peak retention time of volatile organic compounds provided in an embodiment of the present application; Figure 1 As shown, the following steps are included: Step S101, obtaining gas chromatography data of organic compounds at the current moment and historical data of organic compounds; the historical data includes: retention time data of each chromatographic peak of a substance in the gas chromatography measured in the past time period; Step S102, input the gas chromatography data and historical data into a prediction model to predict the retention time of each chromatographic peak of the organic compound at the current moment; the prediction model includes: a first model and a second model; the first model is used to obtain a first prediction value of the retention time based on the gas chromatography data; the second model is used to obtain a second prediction value of the retention time based on the historical data and the first prediction value, and the second prediction value is the retention time predicted by the prediction model; or the second model is used to obtain a second prediction value of the retention time based on the historical data, and the retention time predicted by the prediction model is obtained by fusing the first prediction value and the second prediction value.

[0041] It should be noted that at each monitoring station, the original spectra of the volatile organic compounds monitored are referenced to Figure 2 shown. Figure 2 The medium blue labels are artificially calibrated substances; among them, 1 is ethane, 2 is ethylene, 3 is methane, 4 is propylene, 5 is isobutane, 6 is n-butane, 7 is acetylene, 8 is trans-2-butene, 9 is 1-butene, 10 isobutene, 11 is cis-2-butene, 12 is cyclopentane, 13 is isopentane, and 14 is n-pentane.

[0042] As a further example, the above prediction model is trained as a whole. At this time, the architecture of the prediction model can be seen in Figures 3(b) and 3(c); or when the second model obtains the second prediction value of the retention time based on the historical data, the first model and the second model in the prediction model are trained separately, and then the prediction model is trained, as shown in Figure 3(a); when the second model obtains the second prediction value of the retention time based on the historical data and the first prediction value, the first model in the prediction model is trained separately, and then the prediction model is trained, as shown in Figure 3(d). As shown in Figures 3(b) and 3(d), the first model obtains the first predicted value of the retention time based on the gas chromatography data, and the second model obtains the second predicted value of the retention time based on the historical data and the first predicted value. The second predicted value corresponds to the retention time predicted by the prediction model. As shown in Figures 3(a) and 3(c), the first model obtains the first predicted value of the retention time based on the gas chromatography data, and the second model obtains the second predicted value of the retention time based on the historical data. The retention time predicted by the prediction model is obtained by fusing the first predicted value and the second predicted value.

[0043] Those skilled in the art can understand that the prediction model can be called a dual-modal deep learning model according to its function; the first module can be called an image processing layer (IPLs) according to its function, and the second module can be called a sequence processing layer (SPLs) according to its function. Therefore, in Figures 3(a) to 3(d), the first model and the second model are shown as IPLs and SPLs respectively.

[0044] It should be noted that the prediction models obtained by different training methods (models corresponding to Figures 3(a) to 3(d)) have been verified to have performance that can meet the expected use. Specific performance indicators: Mean Absolute Error (MAE) in retention time positioning can be seen in Table 1: Table 1

[0045] As shown in Table 1, it can be seen that the MAE of retention time prediction using the four prediction model architectures given in Figures 3(a) to 3(d) above is small, can meet expectations, and has good application prospects.

[0046] By further example, since the gas chromatography data is one-dimensional data, the first model can use a one-dimensional convolution kernel to adapt to the input data dimension to achieve retention time prediction based on the gas chromatography data.

[0047] Further preferably, the first model predicts the retention time of each material chromatographic peak based on the gas chromatography data and the derivative data of the gas chromatography data corresponding to the spectrum to obtain a first prediction value; preferably, the derivative includes: a first-order derivative and a second-order derivative. It is understandable that when the retention time is predicted based on gas chromatography, not only the spectrum data of the chromatogram is considered, but also the derivatives at various locations in the spectrum are considered. The retention time prediction is combined with the spectrum data and the derivative (first-order derivative and second-order derivative) characteristics to improve the accuracy of the prediction. Usually, the first-order derivative at the peak of the material chromatographic peak is 0, and the second-order derivative is less than 0. It can be seen that the first-order derivative and the second-order derivative can reflect the characteristics of a certain material chromatographic peak.

[0048] The following example conducts a comparative experiment on the MAE of the first model inputting different data types for retention time prediction, and the results are shown in Table 2: Table 2

[0049] It can be seen that when the first model selects the residual network (ResidualNetwork, ResNet), and the input data of the first model includes graph data, first-order derivatives and second-order derivatives, the MAE of the first model is the lowest.

[0050] It can be understood that, by comparing the data in Table 1 and Table 2, the lowest MAE level is 0.0668 when the single first model is used to predict the retention time, while the MAE of the multimodal prediction model is much less than half of 0.0668 when the retention time is predicted. This proves that the prediction accuracy of the multimodal prediction model provided in this application for retention time prediction is much higher than that of a single modality prediction method.

[0051] More preferably, the above Figure 1 The provided method also includes the following steps: determining the first-order derivative of the retention time of the material chromatographic peak predicted by the prediction model on the corresponding spectrum of the gas chromatography data; correcting the retention time predicted by the prediction model according to the first-order derivative until the corresponding first-order derivative is 0, or until the corresponding first-order derivative is 0 and the second-order derivative is less than 0.

[0052] It should be noted that, in theory, the predicted retention time should be the peak top of the corresponding substance chromatographic peak, but there may be a certain deviation. Therefore, the first-order derivative and / or second-order derivative can be used to search, verify and adjust, see Figure 4 As shown, it is ensured that the final retention time corresponds to the peak top of the substance chromatographic peak.

[0053] Specifically, as an example, the first model mentioned above can be a convolutional neural network; and / or the second model mentioned above can be a recurrent neural network or an attention mechanism type neural network.

[0054] In one embodiment of the present application, the image processing layer adopts a convolutional network structure for one-dimensional sequences to specifically process gas chromatography spectra.

[0055] In one embodiment of the present application, the sequence processing layer adopts a recurrent neural network structure or a neural network structure based on an attention mechanism, which is used to specifically process historical retention time series.

[0056] In one embodiment of the present application, the calculation results of the sequence processing layer and the image processing layer will be aggregated and output after calculation by the fully connected layer, and the output is the final output result of the bimodal deep learning model.

[0057] In one embodiment of the present application, the gas chromatogram is preprocessed, and the preprocessing includes smoothing and derivative generation. The smoothing process uses wavelet analysis and Savitzky-Golay (SG) smoothing algorithm to effectively reduce the noise in the spectrum. The derivative generation calculates the first and second order derivatives of the original spectrum as additional channels and the original spectrum. Figure 1 The image processing layer is input, which can enhance the characteristics of the chromatographic peaks and help the model locate the peak positions more accurately.

[0058] In one embodiment of the present application, the retention time predicted by the dual-modal deep learning model is corrected using a derivative rise algorithm. The algorithm uses the geometric characteristics of the retention time position in the chromatogram, that is, the retention time must correspond to the top of the chromatographic peak, where the first-order derivative of the peak is zero and the second-order derivative is less than zero. In this way, the accuracy of the prediction can be further improved, especially for weak or overlapping chromatographic peaks.

[0059] In one embodiment of the present application, the historical retention time series includes data of several historical time steps. This length is obtained through experimental optimization and can provide sufficient historical information to capture temporal dynamic characteristics while avoiding the introduction of too much redundant information that increases the computational complexity.

[0060] In a more specific embodiment, the method for predicting the retention time of volatile organic compounds provided in the present application comprises the following steps: obtaining a gas chromatogram and preprocessing it to form a multi-channel data feature, and simultaneously obtaining a historical retention time series, and inputting these two types of data into a bimodal deep learning model. Within the bimodal deep learning model, the multi-channel data features are processed by image processing layers (IPLs), and the historical retention time series are processed by sequence processing layers (SPLs). The processed features are combined within the model, and the predicted retention time is output after passing through the fully connected layer. Finally, the derivative ascent algorithm can be applied to the prediction result for further correction to obtain the final retention time prediction value. This method can realize high-precision automated analysis of VOCs, significantly improving the analysis efficiency and accuracy.

[0061] Since gas chromatography data has the dual characteristics of images and time series, this application uses a dual-modal deep learning method to process image features and sequence features separately within the model, which can achieve high-precision analysis of complex industrial-grade continuous monitoring data and effectively solve the difficulties of traditional methods in processing abnormal chromatograms and weak chromatographic peaks. This method can be widely used in VOCs analysis in the fields of environmental monitoring, industrial process control, etc.

[0062] For example, specifically, this embodiment may include the following steps: S1. Data acquisition and preprocessing S11. Obtain a gas chromatogram.

[0063] S12, obtaining a historical retention time series. For each preset moment, extract the historical retention time data of a number of time steps before the moment to form a historical retention time series.

[0064] S13, preprocessing the gas chromatogram. The preprocessing includes smoothing and derivative generation. Smoothing uses wavelet analysis or SG smoothing algorithm, which can effectively reduce the noise in the spectrum. Derivative generation calculates the first and second order derivatives of the original spectrum. The original spectrum, the first order derivative and the second order derivative are used as multi-channel data features.

[0065] S2. Model structure S21, Image Processing Layers (IPLs). IPLs uses a convolutional neural network structure for one-dimensional sequences to process multi-channel data features.

[0066] S22, sequence processing layers (SPLs). SPLs use a GRU network structure to process historically preserved time series. The sequence processing layer uses a recurrent neural network structure or a neural network structure based on an attention mechanism to specifically process historically preserved time series.

[0067] S23. Construct a bimodal deep learning model ResGRU. This model integrates IPLs and SPLs and can process image features and sequence features at the same time. The model first processes the input multi-channel data features and historical retention time series through IPLs and SPLs respectively, then combines the processed features and outputs the final retention time prediction through the fully connected layer.

[0068] S3. Model training and prediction S31. Use the training data set to train the ResGRU model. The training data includes the preprocessed gas chromatogram (multi-channel data features), the historical retention time series, and the corresponding real retention time labels.

[0069] S32. Use the trained ResGRU model to predict the retention time of new gas chromatograms (after preprocessing) and historical retention time series.

[0070] S4. Prediction result correction S41, applying the derivative rise algorithm to correct the predicted retention time obtained in S32. The algorithm uses the geometric characteristics of the RT position in the chromatogram, that is, the RT must correspond to the top of the chromatographic peak, where the first-order derivative of the peak top is zero and the second-order derivative is less than zero.

[0071] S42, iteratively adjusting the predicted point position until a peak position that meets the conditions is found, thereby obtaining a final corrected retention time prediction value.

[0072] S5. Model evaluation and optimization S51. Use the mean absolute error (MAE) as the evaluation indicator to calculate the error between the predicted value and the true value.

[0073] S52. Continuously optimize model performance by adjusting model structure, hyperparameters, etc. until satisfactory prediction accuracy is achieved.

[0074] Preferably, the derivative ascent algorithm described in step S41 is performed according to the following steps: (1) Obtain the RT value initially predicted by the ResGRU model. (2) Calculate the first-order derivative of the point. (3) If the first-order derivative is positive, move the predicted point to the right; if it is negative, move it to the left. (4) Repeat steps (2) and (3) until a point is found where the first-order derivative is zero. (5) Check whether the second-order derivative of the point is less than zero. If so, confirm it as the peak; if not, continue searching.

[0075] It is understandable that the specific structural parameters, training parameters, etc. of the model can be adjusted and optimized according to the actual application scenario. Those skilled in the art can design the corresponding parameters according to actual needs.

[0076] The method of the present application can not only accurately predict the retention time of normal chromatograms, but also, through experiments, it is found that the prediction model provided by the present application can also effectively process abnormal chromatograms due to its high prediction ability, see Figure 5 As shown, Figure 5 (a) shows the weak peak and overlapping peak samples at site D. Figure 5 For the following situations, see (b) for baseline drift, (c) for baseline fluctuation, (d) for retention time drift, and so on. Figure 5 From (a) to (d), it can be seen that the prediction model provided by the present application has a strong resolution capability for chromatograms under unstable and harsh conditions, and can identify chromatographic peaks that are difficult for the human eye to identify under distorted spectrum conditions. In particular, for weak chromatographic peaks, such as olefin compounds such as trans-2-butene, 1-butene, isobutylene and cis-2-butene, the method provided by the present application exhibits excellent recognition capabilities. These compounds have high ozone generation potential and are of great significance to the study of photochemical pollution. Among them, Figure 5 The types of substances are: (i) isobutane; (ii) n-butane; (iii) acetylene; (iv) trans-2-butene; (v) 1-butene; (vi) isobutene; (vii) cyclopentane; (viii) cis-2-butene.

[0077] In addition, the method provided in the embodiment of the present application also has good migration capabilities. Through fine-tuning, the model trained at one monitoring station can be quickly adapted to other monitoring stations, greatly reducing the workload of training the model separately for each station. This makes large-scale industrial applications of VOCs monitoring possible.

[0078] For different monitoring sites, the types and numbers of organic compound combinations monitored may be different. Take sites H, W, D and C as examples, as shown in Table 3: Table 3

[0079] It can be seen that there are inconsistencies in the types of substances monitored between sites, and there are also differences in the order of peak elution of the same substance. This may be due to differences in the use of chromatographic columns, chromatograph method parameter settings, and instrument brands at different sites.

[0080] In one possible implementation, after the prediction model is trained using the training data of the first group of organic compounds: If the prediction model is used to predict the second group of organic compounds, when the second model obtains the second predicted value of the retention time based on the historical data, a small amount of training data of the second group of organic compounds is used to retrain the first model and the second model in the trained prediction model to update the prediction model, and then N dimensional adjustment modules are added to the updated prediction model so that the adjusted prediction model can be used to predict the second group of organic compounds; the number and / or type of material spectra included in the first group of organic compounds and the second group of organic compounds are different, N is an integer greater than or equal to 0, and when the number of the first group of organic compounds and the second group of organic compounds is the same, N is 0; If the prediction model is used to predict the second group of organic compounds, when the second model obtains the second prediction value of the retention time based on the historical data and the first prediction value, a small amount of training data of the second group of organic compounds is used to re-train the first model in the trained prediction model to update the prediction model, and then N dimensional adjustment modules are added to the updated prediction model so that the adjusted prediction model can be used to predict the second group of organic compounds; the number and / or type of material spectra included in the first group of organic compounds and the second group of organic compounds are different, N is an integer greater than or equal to 0, and when the number of the first group of organic compounds and the second group of organic compounds is the same, N is 0.

[0081] It should be noted that the key means of migrating the above prediction model between different sites (different groups of organic compounds) include: 1. Using a small amount of training data to train the first model and / or the second model to adapt it to the new type of material chromatographic peaks; 2. Using the dimension adjustment module to adjust the dimension of the first model and / or the second model and the input and / or output of the data fusion module. If the original first model and / or the second model outputs the prediction results of the retention time of the material chromatographic peaks in N, the number of output results is N. If the number of chromatographic peaks of organic compounds in the new site is M, when M is not equal to N, the output results need to be adjusted. The above-mentioned dimension adjustment model is used to adjust the number of output results. It can be achieved through a fully connected layer.

[0082] See also Figure 6 As shown, the prediction model corresponding to the first group of organic compounds before migration can be understood as the source domain model. When migrating to the second group of organic compounds for monitoring, the prediction model used by the second group of organic compounds can be understood as the target domain model. Taking the prediction model provided in Figure 3(b) as an example, after the first module is trained with a small amount of gas chromatograms of the second group of organic compounds, after predicting the first module, a fully connected layer is added to the output of the first module, and then a fully connected layer is added before the prediction result of the first module and the historical retention time are input into the second module. Finally, a fully connected layer is added to the output of the second module to realize the migration of the prediction model.

[0083] It can be understood that the prediction model provided in the present application takes into account the type and / or quantity of organic compound combinations at different sites and only makes a small adaptive migration of the prediction model, that is, it can effectively migrate it to the new site based on the original prediction model, thereby improving the migration performance of the prediction model and increasing the practicality of the prediction model.

[0084] Figure 6 The source domain data dimension (N, 50, 14) indicates that the source site needs to monitor 14 types of data, and the historical data of the previous 50 hours are used to predict the historical data of the next hour; the target domain data dimension (N, 50, 13) indicates that the target site needs to monitor 13 types of data, and the historical data of the previous 50 hours are used to predict the historical data of the next hour. N represents the number of training samples input into the model during training; the fully connected layer at the output end of the first model is used to convert the 14-dimensional data it outputs into 13 dimensions; the fully connected layer before inputting the second model is used to convert the 13-dimensional data into 14 dimensions; the fully connected layer at the output end of the second model is used to convert the 14-dimensional data into 13 dimensions.

[0085] In summary, the method of this application not only improves the accuracy and efficiency of VOCs monitoring, but also provides more accurate data support for the study of VOCs pollution mechanism over a long time scale, and has broad industrial application prospects. By accurately identifying and quantifying weak chromatographic peaks, especially those olefin compounds with high ozone generation potential, this method provides important support for in-depth research on the mechanism of ozone formation and the causes of photochemical pollution.

[0086] In a specific embodiment, the method provided by the present application is used for continuous monitoring of site H. The monitoring station is classified as a residential area monitoring station, the monitoring frequency is once an hour, and the continuous monitoring time is four months.

[0087] For the data at each preset moment, the present application obtains the following two types of information: (1) gas chromatogram (2) historical retention time series; the gas chromatogram is preprocessed, including smoothing and derivative generation. The smoothing process uses wavelet analysis and SG smoothing algorithm, which can effectively reduce the noise in the spectrum. The derivative generation calculates the first-order and second-order derivatives of the original spectrum. The original spectrum, first-order derivative and second-order derivative are used as multi-channel data features.

[0088] For the historical retention time series, this application extracts the historical retention time data of the 50 time steps before each preset moment.

[0089] This case uses a bimodal deep learning model (abbreviated as ResGRU), and its structure is shown in Figure 3(b). The model consists of the following main parts: (1) Image processing layers (IPLs): using the ResNet network structure to process multi-channel data features. (2) Sequence processing layers (SPLs): using the GRU network structure to process historical retention time series. (3) Fully connected layer: used to combine the outputs of IPLs and SPLs to generate the final retention time prediction.

[0090] When training the model, the preprocessed multi-channel data features and historical retention time series are input into the ResGRU model for training. 90% of the training data is used for training and 10% for validation. The mean absolute error (MAE) is used as the loss function. The training is terminated when the loss function no longer changes significantly.

[0091] The trained ResGRU model is used to predict the validation set data. The derivative ascent algorithm is applied to the prediction results for further correction.

[0092] On the validation set of site H, the mean absolute error (MAE) of the ResGRU model in retention time localization reaches 0.0144 minutes, as shown in Table 4, which is significantly better than the traditional method.

[0093] Table 4

[0094] To verify the migration capability of the model, this application applies the trained model to the monitoring station data of site W, site D, and site C. The migration process is as follows Figure 6 As shown, it mainly includes the following steps: (1) The ResGRU model structure is adjusted so that the outputs of the ResNet and GRU layers are processed through dense layers to adapt to the number of material types at different sites.

[0095] (2) The ResNet layer parameters are fixed and only the GRU layer and dense layer are fine-tuned.

[0096] (3) Use limited data (maybe 160) of the target site for fine-tuning training.

[0097] It was verified that the migrated model can still maintain good performance even when the amount of training data is reduced to 160. For most sites (H, W, D), although the prediction performance of the migrated model is not as good as the locally trained model, it still reaches acceptable engineering standards. In particular, for site C, the performance of all migrated models exceeds that of the locally trained model, showing the great potential of transfer learning in specific sites, as shown in Table 5.

[0098] Table 5

[0099] Figure 7 is a diagram of the system architecture for predicting chromatographic peak retention time of volatile organic compounds provided in an embodiment of the present application, such as Figure 7 As shown, including: The data acquisition module 710 is used to acquire the gas chromatography data of the organic compound at the current moment and the historical data of the organic compound; the historical data includes: the retention time data of each substance chromatographic peak in the gas chromatography measured in the past time period; The retention time prediction module 720 is used to input the gas chromatography data and historical data into a prediction model to predict the retention time of each chromatographic peak of the organic compound at the current moment; the prediction model includes: a first model and a second model; the first model is used to obtain a first prediction value of the retention time based on the gas chromatography data; the second model is used to obtain a second prediction value of the retention time based on the historical data and the first prediction value, and the second prediction value is the retention time predicted by the prediction model; or the second model is used to obtain a second prediction value of the retention time based on the historical data, and the retention time predicted by the prediction model is obtained by fusing the first prediction value and the second prediction value.

[0100] It should be understood that the above-mentioned system is used to execute the methods in the above-mentioned embodiments. The implementation principles and technical effects of the corresponding program modules in the system are similar to those described in the above-mentioned methods. The working process of the system can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0101] Based on the method in the above embodiment, the embodiment of the present application provides an electronic device, such as Figure 8As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the method in the above embodiment.

[0102] Further preferably, the electronic device may be an industrial-grade computer or a dedicated chromatography data analysis device.

[0103] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application.

[0104] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0105] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0106] It is understandable that the processor in the embodiment of the present application may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0107] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, and the software modules can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an ASIC.

[0108] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions may be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0109] It should be understood that the various numerical numbers involved in the embodiments of the present application are only used for the convenience of description and are not used to limit the scope of the embodiments of the present application.

[0110] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A method for predicting the retention time of chromatographic peaks of volatile organic compounds, characterized in that: include: Acquire the gas chromatography data of the organic compound at the current moment and the historical data of the organic compound; the historical data includes: the retention time data of each chromatographic peak of the substance in the gas chromatography measured in the past time period; The gas chromatography data and historical data are input into a prediction model to predict the retention time of each chromatographic peak of the organic compound at the current moment; the prediction model comprises: a first model and a second model; the first model is used to obtain a first prediction value of the retention time based on the gas chromatography data; the second model is used to obtain a second prediction value of the retention time based on the historical data and the first prediction value, and the second prediction value is the retention time predicted by the prediction model; or the second model is used to obtain a second prediction value of the retention time based on the historical data, and the retention time predicted by the prediction model is obtained by fusing the first prediction value and the second prediction value.

2. The method according to claim 1, characterized in that The first model predicts a first predicted value of retention time based on gas chromatography data, comprising: The first model predicts the retention time of each substance chromatographic peak based on the gas chromatography data and the derivative data of the spectrum corresponding to the gas chromatography data to obtain a first prediction value; preferably, the derivative includes: a first-order derivative and a second-order derivative.

3. The method according to claim 1, characterized in that: Also includes: Determining the first-order derivative of the retention time of the substance chromatographic peak predicted by the prediction model on the spectrum corresponding to the gas chromatography data; The retention time predicted by the prediction model is corrected according to the first-order derivative until the corresponding first-order derivative is 0, or until the corresponding first-order derivative is 0 and the second-order derivative is less than 0.

4. The method according to claim 1, characterized in that The first model uses a one-dimensional convolution kernel.

5. The method according to claim 1, characterized in that The prediction model is trained as a whole; or When the second model obtains the second prediction value of the retention time based on the historical data, the first model and the second model in the prediction model are trained separately, and then the prediction model is trained; when the second model obtains the second prediction value of the retention time based on the historical data and the first prediction value, the first model in the prediction model is trained separately, and then the prediction model is trained.

6. The method according to claim 1, characterized in that When the prediction model is trained using the training data of the first group of organic compounds: If the prediction model is used to predict the second group of organic compounds, when the second model obtains the second predicted value of the retention time based on the historical data, a small amount of training data of the second group of organic compounds is used to retrain the first model and the second model in the trained prediction model to update the prediction model, and then N dimensional adjustment modules are added to the updated prediction model so that the adjusted prediction model can be used to predict the second group of organic compounds; the number and / or type of material spectra included in the first group of organic compounds and the second group of organic compounds are different, N is an integer greater than or equal to 0, and when the number of the first group of organic compounds and the second group of organic compounds is the same, N is 0; If the prediction model is used to predict the second group of organic compounds, when the second model obtains the second prediction value of the retention time based on the historical data and the first prediction value, a small amount of training data of the second group of organic compounds is used to re-train the first model in the trained prediction model to update the prediction model, and then N dimensional adjustment modules are added to the updated prediction model so that the adjusted prediction model can be used to predict the second group of organic compounds; the number and / or type of material spectra included in the first group of organic compounds and the second group of organic compounds are different, N is an integer greater than or equal to 0, and when the number of the first group of organic compounds and the second group of organic compounds is the same, N is 0.

7. The method according to any one of claims 1 to 6, characterized in that: The first model is a convolutional neural network; and / or the second model is a recurrent neural network or an attention mechanism type neural network.

8. A prediction system for chromatographic peak retention time of volatile organic compounds, characterized in that: include: A data acquisition module, used to acquire gas chromatography data of organic compounds at the current moment and historical data of organic compounds; the historical data includes: retention time data of each chromatographic peak of a substance in the gas chromatography measured in the past time period; A retention time prediction module is used to input the gas chromatography data and historical data into a prediction model to predict the retention time of each chromatographic peak of the organic compound at the current moment; the prediction model includes: a first model and a second model; the first model is used to obtain a first prediction value of the retention time based on the gas chromatography data; the second model is used to obtain a second prediction value of the retention time based on the historical data and the first prediction value, and the second prediction value is the retention time predicted by the prediction model; or the second model is used to obtain a second prediction value of the retention time based on the historical data, and the retention time predicted by the prediction model is obtained by fusing the first prediction value and the second prediction value.

9. The system according to claim 8, characterized in that The first model used in the retention time prediction module predicts the retention time of each substance chromatographic peak based on the gas chromatography data and the derivative data of the spectrum corresponding to the gas chromatography data to obtain a first prediction value; preferably, the derivative includes: a first-order derivative and a second-order derivative.

10. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method as claimed in any one of claims 1 to 7.

Images