A VOC odor rating model and optimization method

Abstract

The present invention discloses a VOC odor rating model and optimization method. The VOC odor rating model is established and trained by obtaining mass spectrometry data of VOC gas emitted by parts and odor ratings of each part by olfactory engineers. The contribution of gases of different molecular weights to the odor rating is studied, and the molecular weights with greater contribution to the odor rating are screened out. The concentration of a portion of the gases is reasonably reduced by a target optimization method to achieve overall VOC odor rating optimization in the vehicle. The present invention does not require chemical analysis and identification of specific components of VOC gas. It only needs to provide mass spectrometry data to predict the odor rating through the model. In actual production, the VOC odor rating model established by the present invention can quickly determine whether the odor rating of the part is qualified through the mass spectrometry of the parts in the vehicle, without the need for olfactory engineers to judge the odor, thereby improving efficiency and saving labor.

Description

Technical Field

[0001] The present invention specifically relates to a VOC odor rating model and an optimization method. Background Art

[0002] Volatile Organic Compounds (VOC) are organic compounds with a boiling point between 50-250 degrees Celsius that exist in the air in gaseous form at room temperature. Most VOC components have an unpleasant odor, and long-term exposure to this environment will cause harm to human health. According to laboratory verification, the source of odor in the car is mainly volatile organic compounds (VOCs) brought by interior accessories and parts of the car. Considering user health and user experience, it is necessary to analyze the VOC gas in the car and optimize its odor.

[0003] Since VOC refers to a very wide range of organic compound molecules and has a very complex composition, it is difficult to determine which specific volatile organic compound component is the source of the odor in actual research. Current research mainly combines sampling and analysis methods to analyze VOC components. For example, solid phase microextraction (SPME) is a sampling method that can be used to evaluate and locate the presence of VOCs. SPME fibers are used to extract samples from solid or liquid media and separate analytes for analysis. Another example of sample preparation is needle trap microextraction (NTME), which uses a needle trap as a new tool to collect samples and requires a very small sample volume without compromising sensitivity. After sample preparation, the sample needs to be analyzed to obtain more detailed information about the molecules and concentrations. Gas chromatography-mass spectrometry (GC-MS) is the most commonly used method for analyzing VOC components based on different molecular masses and adsorption capacities. The electronic probe contains a sensor array that is used to detect the special characteristics of each VOC component. The results show that both GC-MS and electronic probes can well distinguish different VOC components and quantify their concentrations. Proton transfer reaction mass spectrometry (PTR-MS) is also an analytical method that determines the detailed VOC components by identifying the mass resolution fingerprint of each VOC component. Through these sampling and analysis methods, laboratories are able to determine the specific components of VOC gas and analyze the odor performance of VOC gas based on the physical properties of each component.

[0004] Existing research on VOC gas composition based on sampling and analysis methods generally requires various chemical experiments. For example, sampling and chromatographic analysis require very delicate chemical operations. Since VOC gas is usually composed of many kinds of organic compound molecules, the techniques such as substance separation and identification involved in the experiment are very complex. Given that the actual automobile production process requires simple and rapid detection of VOC gas, this complex and time-consuming sampling and analysis procedure is not practical. Summary of the invention

[0005] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a VOC odor rating model and optimization method.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A VOC odor rating model and optimization method, including the following methods:

[0008] (1) Construct a VOC odor rating model and train the VOC odor rating model;

[0009] (2) Optimize VOC odor rating.

[0010] There is no temporal order between these two methods. When optimizing the VOC odor rating, the optimization method adopted can be applied to the VOC odor rating model constructed in method (1) of the present invention, and can also be applied to other forms of odor rating models.

[0011] Furthermore, the method for constructing the VOC odor rating model includes the following steps:

[0012] (1) Process the original data;

[0013] The raw data includes instrument data and manual scoring data; the instrument data is data obtained by mass spectrometry analysis of VOC gases emitted by vehicle parts by a mass spectrometer; the manual scoring data refers to the odor rating given by olfactory engineers to VOC gases emitted by vehicle parts based on their own subjective feelings;

[0014] (2) Determine the VOC odor rating model;

[0015] According to the characteristics of the original data, it is determined that the model describes the relationship between physical measurement quantities and human perception quantities, so the VOC odor rating model is determined as:

[0016]

[0017] in, is the predicted odor rating of the nth sample part, β m is the regression coefficient on the mth related molecular weight, z n，m is the signal value of the nth sample at the mth molecular weight, β′0 is a constant term, and M is the total number of molecular weights used in the model.

[0018] (3) Train the model;

[0019] The signal data z of the nth sample part n，m and odor rating Substituting into formula (2), we can find the optimal β = [β1, ..., β M ,β′0] parameter combination to minimize the loss function.

[0020] Furthermore, the specific process of data processing is:

[0021] (1.1) Select the data obtained by Hg ion source bombardment;

[0022] (1.2) Select the molecular weight of the VOC gas component that may be in the vehicle and the corresponding signal data;

[0023] (1.3) Compare the signal data obtained in step (1.2) with the nitrogen background signal data, remove the noise signal, and finally obtain the data for modeling.

[0024] Furthermore, during the model training process, the Lasso regression method is used, that is, the loss function with the L1 norm regularization term ||β||1 is added. The specific formula is as follows:

[0025]

[0026] Where N represents the total number of samples used in the model training process. N samples correspond to N groups of data. n represents the nth sample (n = 1, 2, ... N). n represents the true odor rating of the nth sample, is the smell rating predicted by the model for the nth sample, and γ is a constant coefficient used to control the impact of the L1 norm term on the entire loss function.

[0027] Further, the optimization method of VOC odor rating includes the following steps:

[0028] (1) Screen out the molecular weights that contribute most to the odor rating;

[0029] (2) Determine the optimization plan to achieve optimization of odor rating.

[0030] Furthermore, the specific process of step (1) of the VOC odor rating optimization method is as follows:

[0031] (1) Find typical samples: Substitute all sample data into the same model and observe the molecular weight distribution screened by the model; if a large number of samples can obtain a consistent molecular weight contribution distribution, one or several samples can be used as typical samples, and the results can represent the molecular weight screening results of a large number of samples on the model;

[0032] (2) After determining the typical samples, screen the molecular weights that contribute most to the odor rating;

[0033] The molecular weight contribution is determined according to the following criteria:

[0034] 2.1) For typical samples, how many models select a certain molecular weight as the molecular weight with the largest contribution; if a molecular weight is selected by a model, the number of times the molecular weight is selected by the model is increased by one. The more times a molecular weight is selected, the more models select the molecular weight, and the higher the credibility of its contribution;

[0035] 2.2) For typical samples, when a molecular weight is selected by a model, if the molecular weight has a higher contribution ranking on the model, it is given a larger weight; if the molecular weight has a lower contribution ranking on the model, it is given a smaller weight. The weighted scores of each molecular weight on all models are summed to obtain the total score of the molecular weight. The total scores are sorted from large to small to obtain the final result of the contribution of the molecular weight in the model. The above weighted score summation process is expressed by formula (5):

[0036]

[0037] Among them, Q m is a subset, indicating the number of the model selected from the K models with the mth molecular weight, f K represents the Kth model, v q，j It means that the mth molecular weight is ranked j according to its contribution in the qth selected model, and the corresponding weighted score v is obtained q，j , sum up all weighted scores to obtain the total score g of the mth molecular weight m , according to all g m The molecular weight contribution can be ranked, and the molecular weight screening results can be selected from the top molecular weights in the ranking according to needs.

[0038] Furthermore, the specific process of step (2) of the VOC odor rating optimization method is as follows:

[0039] Substitute the mathematical model between the signal and the odor rating at each molecular weight and the screened molecular weight set into formula (6) to find the best solution for optimizing the odor rating;

[0040]

[0041]

[0042] The set I represents the molecular weight set used for optimization, which is a subset of the molecular weights with greater contribution that have been screened out. M is the number of all molecular weights used in the model. f represents the VOC odor rating model. z m 、z iare respectively the mass spectrometry signal data of the optimized sample parts at the corresponding molecular weights, m and i refer to the mth and ith molecular weights of the optimized sample parts respectively, i is in set I, and m is not in set I; is the positive deviation between the actual mass spectrometry signal data and the target mass spectrometry signal data, is the negative deviation between the actual mass spectrometry signal data and the target mass spectrometry signal data, y goal Rating the optimization goal you want to achieve, w i The difficulty of adjusting the original data at the corresponding molecular weight is defined as i The value of is between 1 and 5. The closer the difficulty is to 1, the easier the actual operation is. The closer the difficulty is to 5, the more complicated the actual operation is. i The size of w is adjusted first during the optimization process. i Data on smaller molecular weights.

[0043] The beneficial effects of the present invention are:

[0044] (1) The present invention establishes a VOC odor rating model, which describes the relationship between mass spectrum data and VOC odor rating. There is no need to chemically analyze and identify the specific components of the VOC gas. Only the mass spectrum data is required to predict the odor rating through the model. In actual production, the model can quickly determine whether the odor rating of the part is qualified through the mass spectrum of the vehicle parts, without the need for olfactory engineers to judge the odor, thereby improving efficiency and saving labor. In addition, from the model established by the present invention, the contribution of each molecular weight corresponding to the substance to the odor rating can be quantified, and then the VOC components with greater contributions can be selected to prioritize the design of odor optimization solutions.

[0045] (2) The weighted ranking molecular weight screening method based on typical samples and multiple models of the present invention can scientifically and efficiently screen out the molecular weights that have a greater impact on the odor rating. In actual production, it can be used as a specific direction for optimizing the odor in the car and guide production.

[0046] (3) The present invention can provide specific optimization solutions for different gas combinations through the target optimization algorithm, so that the odor in the car is improved to the target rating. The optimization method of the present invention is convenient and fast to operate, highly efficient and flexible. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is an example of a raw data table.

[0048] Figure 2 This is a comparison of the nitrogen background 3σ confidence interval and the sample signal.

[0049] Figure 3It is a flow chart of the method for constructing a VOC odor rating model of the present invention.

[0050] Figure 4 are the model parameters β1, β2, ...β 64 Distribution map of .

[0051] Figure 5 This is the molecular weight-contribution distribution diagram of sample No. 60 on this model.

[0052] Figure 6 is the molecular weight-contribution plot for all samples on the same model.

[0053] Figure 7 This is the molecular weight-contribution relationship diagram of sample No. 60 with a score of 3 substituted into 20 models. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings. It should be pointed out that the specific implementation methods are only detailed descriptions of the present invention and should not be regarded as limitations of the present invention.

[0055] Example 1

[0056] A method for constructing a VOC odor rating model includes the following steps:

[0057] (1) Processing of raw data:

[0058] The raw data includes instrument data and manual scoring data; the instrument data is data obtained by mass spectrometry analysis of VOC gases emitted by vehicle parts by a mass spectrometer; the manual scoring data refers to the odor rating given by olfactory engineers to VOC gases emitted by vehicle parts based on their own subjective feelings;

[0059] The data processing process includes:

[0060] (1.1) Select the data obtained by Hg ion source bombardment;

[0061] (1.2) Selecting the molecular weights and corresponding signal data that may be the VOC gas components in the vehicle; in specific implementation, searching for chemical substances that may be the VOC gas components in the vehicle in literature or related materials, calculating the molecular weights of these chemical substances, and finally obtaining the molecular weights, and selecting these molecular weights and corresponding signal data for processing.

[0062] (1.3) Compare the signal data obtained in step (1.2) with the nitrogen background signal data, remove the noise signal, and finally obtain the data for modeling.

[0063] (2) Determine the VOC odor rating model:

[0064] In the process of model building, based on the characteristics of the original data, it was determined that the model needed to describe the relationship between the physical measurement quantity and the human perception quantity, and the VOC odor rating model was determined with reference to the Weber-Fechner theorem;

[0065] (3) Train the model:

[0066] The instrument data (i.e., mass spectrum data) is used as the input of the model, and the manual scoring data (i.e., odor rating data) is used as the output of the model, and a large number of sample parts are measured; the processed data is substituted into the model for model training; in order to obtain more accurate and reasonable model parameters, the Lasso regression method is used in model training to determine the model parameters and obtain a complete mathematical model.

[0067] (1) Processing the original data

[0068] In some preferred embodiments, the instrument data is the data measured by a mass spectrometer, and the working principle of the mass spectrometer is: the gas molecules introduced into the mass spectrometer are bombarded with three ion sources, Hg, Xe, and He, so that the gas molecules lose an electron, while the molecular weight remains unchanged. Since the deflection of the gas molecule ions that lose electrons in the mass spectrometer is only related to their molecular weight, they can be separated according to different molecular weights. Eventually, the gas molecule ions of the same molecular weight arrive at a fixed position at the receiving end of the mass spectrometer, and the signal received at this position can be counted to count the material components corresponding to the molecular weight.

[0069] In some preferred embodiments, the ratings and standard descriptions given by olfactory engineers to odors are specifically shown in Table 1:

[0070] Table 1 Odor rating standards

[0071]

[0072]

[0073] *When the odor is between two rating levels, a rating of 0.5 points is used.

[0074] It can be seen from Table 1 that the higher the odor rating score, the more unpleasant the odor is.

[0075] Figure 1 The results of the analysis of VOC gases emitted by sample parts in the vehicle derived from the mass spectrometer are given. Figure 1As shown, the top row "score" is the odor rating of the sample parts given by the olfactory engineer, and it can be seen that the odor rating of the car interior parts named "Auria carpet foam 03" is 3 points; in the first column "M+number" on the left, the number represents the value of the molecular weight (massweight), for example, "M17" means the molecular weight 17; the second column is the unit cps (counts per second) of the measurement signal at the molecular weight, which represents the signal intensity received by the mass spectrometer at the molecular weight per second; "*" indicates that the signal is obtained by bombardment of the Xe ion source, if there is no "*", it means that the signal is obtained by bombardment of the Hg ion source; the third column "Background" indicates that this column is the background data signal, and the background data signal is obtained by directly passing nitrogen into the mass spectrometer; the fourth column is the cps signal result obtained when the VOC gas emitted by the car interior parts named "Auria carpet foam 03" is passed into the mass spectrometer.

[0076] Depend on Figure 1 It can be seen that the data measured by the mass spectrometer ranges from 7, 16, 17 to 300 molecular weights, and is divided into the results of bombardment by two ion sources, Xe and Hg. Since the ionization energy of the chemical components needs to be considered during ion source bombardment, one is whether the gas molecules can be effectively bombarded to lose an electron, and the other is to avoid excessive energy causing the gas molecules to fragment and change the molecular weight. Taking the above two points into consideration, the mass spectrum data obtained by bombarding the gas molecules with the Hg ion source is selected as the data for modeling from the Xe and Hg ion source data provided by the mass spectrometer. At a molecular weight of 30, since Hg ions have no signal, the results of Xe ions are used instead.

[0077] After determining to use Hg ion source bombardment to obtain data, based on the common chemical components of VOC gas in the car, the continuous molecular weight range from 17 to 300 in the data obtained by Hg ion source bombardment was narrowed down to 64 molecular weights (in this embodiment, the chemical substances that may be the components of VOC gas in the car were searched in the literature, and the molecular weights of these chemical substances were calculated, and finally 64 molecular weights were obtained. These 64 molecular weights and the corresponding signal data were preferably processed); it is believed that the chemical components corresponding to these 64 molecular weights may affect the odor score. Among these 64 molecular weights, 31 molecular weights have isomers, and 33 molecular weights correspond to only one molecular structure.

[0078] After narrowing the molecular weights to 64, the signal from each sample part was compared with the background signal to determine whether the signal was noise. The nitrogen background data was measured multiple times, and the average μ and standard deviation σ of the nitrogen background signal at each molecular weight were obtained.

[0079] In some preferred embodiments, the noise determination criteria are: if the signal of the sample part at a certain molecular weight is within the 3σ (three times standard deviation) confidence interval of the nitrogen background data at the molecular weight, the sample signal is considered as noise removal. The conditions that the noise measurement needs to meet are as follows:

[0080] μ-3σ<noise<μ+3σ

[0081] Figure 2 The average value and 3σ confidence interval of all nitrogen background data are plotted in Figure 1, and the data of a sample part are attached for comparison. If the signal of the sample part falls within the gray range, it is regarded as noise.

[0082] To summarize, in this embodiment, the raw data processing includes three steps. The first step is to select the cps signal data obtained by bombardment of the Hg ion source. The second step is to select 64 molecular weights that may be the VOC gas components in the vehicle from the continuous molecular weight of 17-300 (at a molecular weight of 30, since Hg ions have no signal, the results of Xe ions are used instead). The third step is to compare the cps signals at these molecular weights with the nitrogen background signal data to determine whether the signal is noise that needs to be removed. If so, the noise is removed to obtain data for modeling.

[0083] (2) Determine the VOC odor rating model

[0084] After processing the original data, the sample data used for modeling includes signals at 64 molecular weights and odor ratings given by olfactory engineers. In order to describe the relationship between the original cps signal of the sample and the odor rating, and to analyze the contribution of each molecular weight signal to the odor rating, it is necessary to establish a model with cps signals at 64 molecular weights as input and odor ratings as output. The model parameters are obtained through model training with a large amount of known data, and the quantitative relationship between input and output is given. After obtaining the quantitative relationship between input and output, the model can be used to directly predict the odor rating of the original sample part based on the cps signal data.

[0085] In the process of model building, the Weber-Fechner theorem is used as a reference. This theorem describes the relationship between the intensity of subjective sensations (such as smell, taste, hearing, etc.) experienced by the human body and the intensity of external stimuli. The specific formula is as follows:

[0086] P = klogI

[0087] Wherein, P represents the subjective intensity perceived by the human body, k is a constant, and I is the signal of physical measurement. According to the logarithmic-linear relationship between human perception and physical signal given in the theorem, the model formula in the present invention is determined as follows:

[0088]

[0089] In formula (1), is the predicted odor rating of the nth sample part, β m is the regression coefficient on the mth related molecular weight, z n，m is the cps signal value of the nth sample at the mth molecular weight, s m is the sensitivity coefficient (conversion coefficient) from the original cps signal to concentration at the mth molecular weight, w m is the odor detection threshold of the specific chemical component corresponding to the mth molecular weight, β0 is a constant term, and M is the total number of molecular weights used in the model (in this embodiment, the total number of molecular weights used is 64, ie, M=64).

[0090] Since the olfactory threshold of the isomers of gas molecules corresponding to certain molecular weights is unknown, the conversion coefficients of the mass spectrometer from cps signal to concentration at each molecular weight are also unknown, but these coefficients are constants, so their influence can be removed from the ln term and combined with the constant term β0 of the model (for molecular weights with isomers, it is assumed that there is only one VOC gas). Based on this, formula (1) is processed as follows:

[0091]

[0092] Then we can get formula (2) as follows:

[0093]

[0094] In the improved formula (2), the mass spectrum signal z n，m is the input of the model, the odor rating is the output of the model. is the odor rating predicted by the model for the nth sample, y n is the real odor rating given by the known olfactory engineer. By substituting a large amount of known data into the input, the predicted odor rating of each sample is obtained. The difference between the predicted odor rating and the real odor rating is optimized by combining the model training algorithm to obtain the optimal regression coefficients β1,...,β M , β′0, determine the cps signal intensity z at each molecular weight n，m Rating of odor The quantitative relationship between them.

[0095] (3) Model training process

[0096] After processing the raw data and determining the mathematical model, model training can begin. The specific process is as follows:

[0097] The cps signal data z of the nth sample part n，m and odor rating n Substituting into formula (2), we can find the optimal β = [β1, ..., β M ,β′0] parameter combination can minimize the loss function.

[0098] In this embodiment, due to the large number of sample features, a total of 65 parameters need to be calculated, which easily leads to overfitting. In order to avoid the problem of overfitting, Lasso regression is used, that is, a loss function with an L1 norm regularization term ||β||1 is added. The specific formula is as follows:

[0099]

[0100] Where N represents the total number of samples used in the model training process. N samples correspond to N groups of data. n represents the nth sample (n = 1, 2, ... N). n represents the true odor rating of the nth sample, is the odor rating of the nth sample predicted by the model, and γ is a constant coefficient used to control the influence of the L1 norm term on the entire loss function. In this embodiment, its specific value is set to 0.6.

[0101] By adding the L1 norm regularization term of parameter β, reasonable parameter training results can be obtained.

[0102] Figure 3 All operations of raw data processing, model building, and model training are integrated to provide a complete algorithm flow of the solution of the present invention.

[0103] Specific implementation examples

[0104] The original mass spectrum data and odor ratings of 179 in-vehicle sample parts were obtained from the experiment. The original mass spectrum data table is as follows: Figure 1 As shown, a total of 362 measurements of cps signals on molecular weights bombarded by different ion sources were involved.

[0105] according to Figure 3 The process shown is to read the data in the Excel table and obtain the corresponding cps signal data matrix Odor Rating Data And 23 sets of background data

[0106] Determine the index of the 64 molecular weights used in the model in D0 and B0, extract the data on these molecular weights, and obtain The 23 columns of background cps signals in B are integrated to calculate the overall mean and variance to obtain Compare the 179 columns of data in D with μ B ±3σ B For comparison. If a cps signal is in the corresponding μ B ±3σ B If it is within the range, it is updated to 1. Since the logarithm of 1 is zero, the noise signal can be removed from the model data through this operation.

[0107] Take the ln logarithm of each item in the updated D to obtain Combination We select 149 data sets from 179 data sets as training sets for model training, and the remaining 30 data sets as test sets. D train and S train As the input of the model in formula (2) With output Substituting into the model and training according to the loss function (3) given by Lasso regression, we can obtain the optimal model parameters β = [β1, β2...β 64 , β′0], and finally obtain the exact mathematical model.

[0108] After obtaining the model, 30 sets of test data D test Substitute into the model and calculate the predicted scores of 30 test data sets based on the model parameter β Comparative prediction scores The actual score y n ∈S test , the accuracy of the model can be obtained.

[0109] The 179 data sets were divided into 149 training data sets and 30 test data sets, and a model with an accuracy of 0.93 was obtained. Among the β parameters of this model, the first 64 parameter values ​​are as follows Figure 4 As shown in the figure, the value of the 65th item β′0 is 2.8663. Substituting the data of sample No. 60 into the model, the molecular weight-contribution distribution diagram of the sample can be obtained, as shown in Figure 5 shown.

[0110] Example 2

[0111] This implementation includes two aspects: one is to screen out the molecular weights that contribute more to the odor rating based on the existing VOC mass spectrum data and the mathematical model of odor rating; the other is to select a combination from the screened molecular weights to achieve the optimization of the odor rating.

[0112] (i) Screening is performed based on existing VOC mass spectrum data and a mathematical model for odor rating (the model here may be the model obtained in Example 1 or another form of odor rating model); the influence of data at different molecular weights on odor ratings is also different, and molecular weights with greater contributions are preferably selected.

[0113] Since the establishment of the model is highly dependent on factors such as the amount of original data and data distribution, we will follow the same modeling approach, set up different training sets and test sets, train multiple models, and through comprehensive analysis of the common performance of these multiple models, screen out the molecular weight that contributes most to the odor rating.

[0114] (ii) Based on the VOC odor rating model and the screened molecular weights, a corresponding optimization is established, by adjusting the mass spectrometry data signal size corresponding to these molecular weights so that the adjusted signal can reach the target odor rating through model calculation. The solution that requires the least adjustment to achieve the target is the optimal solution.

[0115] (I) Molecular weight screening for greater contribution to odor rating

[0116] For describing VOC mass spectrum data n，m (z n，m represents the signal data of the nth sample at the mth molecular weight, assuming that a total of N samples and M molecular weights are used) and the odor rating y n (representing the odor rating of the nth sample), and its mathematical model format is defined as:

[0117] y n =f(z n，m ), m=1,2,…,M (4)

[0118] The meaning of the model is: for the nth sample, the model uses mass spectrum data z at several molecular weights n，m As input, the smell rating y of the sample n is the output, and f is the mathematical relationship between the mass spectrum data and the odor rating.

[0119] This model can reflect the impact of mass spectrum data at different molecular weights on the odor rating (defined as the contribution of the molecular weight to the odor rating). Since all molecular weights cannot be considered in actual production, it is necessary to screen out molecular weights with greater contributions for optimization.

[0120] Since the model is directly affected by the training data and test data during the training process, choosing different training sets and test sets may result in significantly different contributions of different molecular weights. Therefore, in order to avoid the particularity of a single model, in the process of molecular weight screening, K groups of different training sets and test sets are randomly defined to obtain K models. The mathematical form of these K models is the same, but the specific model parameters will be different due to the different training and test sets. Finally, there are N groups of sample part data and K models. Comprehensive analysis of the molecular weight contribution performance of these sample data on the model can obtain comprehensive, reliable and representative molecular weight screening results.

[0121] Due to the large amount of sample data and the large number of models, substituting each sample into each model to rank the contribution and then synthesizing the results requires a lot of data processing. Therefore, the weighted ranking method of typical samples-multiple models is adopted. This method can screen efficiently and comprehensively. The specific steps are as follows:

[0122] (1) Find typical samples: Substitute all sample data into the same model and observe the molecular weight distribution screened by the model; if a large number of samples can obtain a consistent molecular weight contribution distribution, one of them can be used as a typical sample, and its result can represent the molecular weight screening results of a large number of samples on this model. Similarly, by summarizing the screening results of one or several typical samples on all models, the screening results of all samples on all models can be obtained.

[0123] (2) After determining the typical samples, screen the molecular weights that contribute most to the odor rating;

[0124] Since K models provide K different molecular weight contribution rankings for each sample, all models are combined to determine the overall molecular weight contribution according to the following two criteria:

[0125] 2.1) For typical samples, how many models select a certain molecular weight as the molecular weight with the largest contribution; if a molecular weight is selected by a model, the number of times the molecular weight is selected by the model is increased by one. The more times a molecular weight is selected, the more models select it, and the higher the credibility of its contribution;

[0126] Considering standard 2.2) on the basis of 2.1), for typical samples in 2.2), when a molecular weight is selected by a model, if the molecular weight has a higher contribution ranking on the model, a larger weight is assigned (for example, the molecular weight with the highest contribution is 12 points); if the molecular weight has a lower contribution ranking on the model, a smaller weight is assigned (for example, the molecular weight with the second highest contribution is 11 points). Then, the weighted scores of each molecular weight on all models are summed to obtain the total score of the molecular weight. The total scores are sorted from large to small to obtain the final result of the contribution of the molecular weight in the model. According to the final result, the molecular weights ranked in the top few positions are selected for priority consideration in the optimization scheme. Generally, it depends on the actual situation. Preferably, the top 12 molecular weights are used. The above weighted score summation process can be expressed by formula (5):

[0127]

[0128] Where Q m is a subset, indicating the number of the model selected from the K models with the mth molecular weight, f K represents the Kth model, v q，j It means that the mth molecular weight is ranked j according to its contribution in the qth selected model, and the corresponding weighted score v is obtained q，j , sum up all weighted scores to obtain the total score g of the mth molecular weight m , according to all g m The molecular weight contributions can be sorted from large to small, and the molecular weight screening results can be selected from the top molecular weights in the sorting according to needs.

[0129] (II) Optimization scheme of odor rating based on mathematical model

[0130] In order to optimize the odor rating in the car, it is necessary to reduce the concentration of VOC components in the car to reduce the odor. Through the previous analysis, the molecular weights that contribute most to the odor rating in the car were selected. It is believed that the VOC components corresponding to these molecular weights have a greater impact on the increase in the odor rating (the higher the odor rating, the more unacceptable the odor). If you want to reduce the odor rating and get a better odor rating, you can give priority to processing the gas components corresponding to these molecular weights.

[0131] The mapping relationship between the original signal and the odor rating at each molecular weight is shown in formula (4). By inputting the signal data at each molecular weight, the model can predict the corresponding odor rating. Then the size of the signal at the molecular weight can be changed so that the odor rating predicted by the model reaches the odor rating set by the optimization target. This optimization problem with a clear known goal belongs to the goal programming problem, and its general form is shown in formula (6):

[0132]

[0133]

[0134] Wherein, set I represents the molecular weight set used for optimization, which is generally a subset of molecular weights with greater contribution that are screened out, M is the number of all molecular weights used in the model, and f represents the VOC odor rating model. The VOC odor rating model here can be the model shown in formula (2) or any other form of model, as long as it is a mathematical model that describes the relationship between VOC mass spectrometry signal data and odor rating; m 、z i They are respectively the mass spectrum signal data of the optimized sample parts at the corresponding molecular weights, m and i respectively refer to the mth and ith molecular weights of the optimized sample parts, i is in set I, and m is not in set I. is the positive deviation between the actual mass spectrometry signal data and the target mass spectrometry signal data, is the negative deviation between the actual mass spectrometry signal data and the target mass spectrometry signal data, y goal Rating the optimization goal you want to achieve, w i The difficulty w is defined as the difficulty of adjusting the raw data at the corresponding molecular weight (equivalent to the difficulty of increasing or decreasing the gas concentration corresponding to the molecular weight). i The value of is between 1 and 5. The closer the difficulty is to 1, the easier the actual operation is. The closer the difficulty is to 5, the more complicated the actual operation is. i The size of w is adjusted first during the optimization process. i The data on molecular weights with smaller values ​​can be used to obtain an optimization plan that is more suitable for actual production.

[0135] For a specific model, after substituting the specific form of the model into f in formula (6), the target optimization problem design of the model can be obtained, and the best solution for odor rating optimization can be obtained by finding the optimal solution.

[0136] Implementation Example 3

[0137] This embodiment adopts the model constructed in Embodiment 1 and the optimization method in Embodiment 2.

[0138] Molecular weight screening:

[0139] (1) Find typical samples; train all 179 samples using 20 different training set and test set division methods to obtain 20 sets of model parameters. Substitute the 179 samples into one of the models to obtain the contribution degree of all samples on the same model - molecular weight. The results are as follows: Figure 6 As shown by Figure 6It can be seen that for the same model, the overlap of the contribution-molecular weight graphs obtained by the model for all sample data is very high, that is, the molecular weights (i.e., peak positions) screened out by the same model for all samples are almost the same. Therefore, among the 179 samples, one sample is randomly selected from the samples with scores of 2.5, 3, 3.5, 4, and 4.5 as a typical sample (in this embodiment, the selected typical samples are samples No. 13, 60, 78, 142, and 49, respectively).

[0140] (2) After determining the typical samples, screen the molecular weight that contributes most to the odor rating:

[0141] The sample data No. 60 with a score of 3 was traversed through 20 models, and 20 sets of molecular weight-contribution results were obtained as follows Figure 7 As shown, each molecular weight-contribution line corresponds to a result given by a model. For each molecular weight-contribution line, the molecular weights are sorted according to the peak values ​​of the contribution, and the top twelve peaks are retained. It is assumed that the corresponding molecular weights are 76, 70, 126, 104, 68, 78, 90, 92, 46, 45, 72, and 144 respectively;

[0142] The molecular weight 76 ranked first in contribution is assigned a weight of 12 points, the molecular weight 70 ranked second in contribution is assigned a weight of 11 points, and so on. Finally, the 20 models generate 20 molecular weight rankings of the top twelve contributions. Assume that for the molecular weight 76, it is selected in each model and is assigned a weighted score of v q，j The weighted scores of the molecular weight 76 in the 20 models are 12, 11, 10, 12, 12, 12, 12, 12, 12, 12, 10, 12, 11, 11, 12, 12, 10, 11, 12, 12, respectively. The sum of the weighted scores of all the selected molecular weights is 230. The sum of the weighted scores is calculated for all the selected molecular weights, and they are sorted by score. The top twelve are taken to obtain the top twelve molecular weights selected according to contribution.

[0143] 2. Odor Rating Optimization Plan:

[0144] After screening out the molecular weights with the largest contribution to the odor rating, the top five molecular weights of contribution, 70, 76, 106, 90, and 104, are selected, and the combination of these five molecular weights is optimized. Substituting the specific logarithmic-linear mathematical model shown in formula (2) and the set I of these five molecular weights into formula (6), the specific optimization problem design in this embodiment can be obtained, as shown in formula (7):

[0145]

[0146]

[0147] In actual production, the difficulty of adjusting the raw data at each molecular weight is determined according to actual operation experience; in specific implementation, the difficulty of operation is set by the user according to the actual situation. In this embodiment, for the purpose of illustration, the difficulty of operation w of five molecular weights is randomly set. i , as shown in Table 2.

[0148] Table 2 Operation difficulty of five molecular weights

[0149] Molecular weight 70 76 106 90 104 Weight 1 4 2 1.3 1.5

[0150] It can be seen from Table 2 that although the molecular weight of 76 ranks high in the contribution ranking and has a large contribution, the actual operation is also difficult. Therefore, the molecular weight of 76 is not the most prioritized molecular weight in the optimization plan.

[0151] Three samples were selected from different odor ratings, namely sample 58 with an odor rating of 3.5, sample 50 with an odor rating of 4.0, and sample 22 with an odor rating of 5.0. The optimization goal was set to reduce the original odor rating by 0.5, 1, and 1.5 points. For example, the model score of sample 58 was reduced from 3.49 to 3.0, 2.5, and 2 points, and the optimization scheme required to achieve these optimization goals was calculated. The optimization scheme is expressed in percentages, which specifically means how much the signal on each molecular weight needs to be reduced to achieve the corresponding optimization goal. The calculated optimization results are shown in Table 3-5 below:

[0152] Table 3 Optimization scheme for sample 58 based on five molecular weight combinations

[0153]

[0154] Table 4 Optimization scheme for sample 50 based on five molecular weight combinations

[0155]

[0156] Table 5 Optimization scheme for sample 22 based on five molecular weight combinations

[0157]

[0158] In Table 3-5, the odor rating is the actual score given by the olfactory engineer to the sample parts, and the model score is the score calculated by substituting the cps signal of the sample into the model, that is, the score given by a model that can predict the odor rating. For the scheme in Table 3-5, taking the optimization of sample 22 odor rating from model score 4.43 to 3.0 as an example, the corresponding optimization scheme can be specifically explained as follows: If you want to optimize sample No. 22 with a model score of 4.43 to an odor rating of 3.0, you need to reduce the signal intensity corresponding to the molecular weight of 70 by 15%, and then reduce the signal intensity corresponding to the molecular weight of 76 by 5%. The signals at the other three molecular weights do not need to be reduced. In actual production, the reduction in signal intensity is converted into a reduction in the concentration of specific gas molecules to optimize the odor rating of VOC gas.

[0159] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

Claims

1. An optimization method for VOC odor rating ， Its characteristics include the following methods: (1) Construct a VOC odor rating model and train the VOC odor rating model; (2) Optimizing the VOC odor rating includes the following steps: 1) Screen out the molecular weight that contributes most to the odor rating. The specific process of step 1) is as follows: I) Find typical samples: Substitute all sample data into the same model and observe the molecular weight distribution screened by the model; if a large number of samples can obtain a consistent molecular weight contribution distribution, one or several samples can be used as typical samples, and the results can represent the molecular weight screening results of a large number of samples on the model; II) After determining the typical samples, screen the molecular weight that contributes most to the odor rating; The molecular weight contribution is determined according to the following criteria: A) For typical samples, how many models select a certain molecular weight as the molecular weight with the largest contribution; if a molecular weight is selected by a model, the number of times the molecular weight is selected by the model is increased by one. The more times a molecular weight is selected, the more models select the molecular weight, and the higher the credibility of its contribution; B) For typical samples, when a molecular weight is selected by a model, if the molecular weight has a higher contribution ranking on the model, it is given a larger weight; if the molecular weight has a lower contribution ranking on the model, it is given a smaller weight. The weighted scores of each molecular weight on all models are summed to obtain the total score of the molecular weight. The total scores are sorted from large to small to obtain the final result of the contribution of the molecular weight in the model. The above weighted score summation process is expressed by formula (5): Among them, Q m is a subset, indicating the number of the model with the mth molecular weight selected from the K models, f K represents the Kth model, v q，j It means that the mth molecular weight is ranked j according to its contribution in the qth selected model, and the corresponding weighted score v is obtained q，j , sum up all weighted scores to obtain the total score g of the mth molecular weight m , according to all g m The molecular weight contribution can be ranked, and the molecular weight screening results can be selected from the top molecular weights in order according to the needs; 2) Determine the optimization scheme to optimize the odor rating. The specific process of step 2) is as follows: Substitute the mathematical model between the signal and the odor rating at each molecular weight and the screened molecular weight set into formula (6) to find the best solution for optimizing the odor rating; The set I represents the molecular weight set used for optimization, which is a subset of the molecular weights with greater contribution that have been screened out. M is the number of all molecular weights used in the model. f represents the VOC odor rating model. z m 、z i are respectively the mass spectrometry signal data of the optimized sample parts at the corresponding molecular weights, m and i refer to the mth and ith molecular weights of the optimized sample parts respectively, i is in set I, and m is not in set I; is the positive deviation between the actual mass spectrometry signal data and the target mass spectrometry signal data, is the negative deviation between the actual mass spectrometry signal data and the target mass spectrometry signal data, y goal Rating the optimization goal you want to achieve, w i The ease of adjusting the raw data to the corresponding molecular weight.

2. A VOC odor rating model according to claim 1, Its characteristics are: The method for constructing the VOC odor rating model includes the following steps: (1) Process the original data; The raw data includes instrument data and manual scoring data; the instrument data is data obtained by mass spectrometry analysis of VOC gases emitted by vehicle parts by a mass spectrometer; the manual scoring data refers to the odor rating given by olfactory engineers to VOC gases emitted by vehicle parts based on their own subjective feelings; (2) Determine the VOC odor rating model; According to the characteristics of the original data, it is determined that the model describes the relationship between physical measurement quantities and human perception quantities, so the VOC odor rating model is determined as: in, is the predicted odor rating of the nth sample part, β m is the regression coefficient on the mth related molecular weight, z n，m is the signal value of the nth sample at the mth molecular weight, β′ 0 is a constant term, and M is the total number of molecular weights used in the model; (3) Train the model; The signal data z of the nth sample part n，m and odor rating Substituting into formula (2), we can find the optimal β = [β 1 , …, β M , β′ 0 ] parameter combination to minimize the loss function.

3. A VOC odor rating model according to claim 2, Its characteristics are: The specific process of data processing is: (1.1) Select the data obtained by Hg ion source bombardment; (1.2) Select the molecular weight of the VOC gas component that may be in the vehicle and the corresponding signal data; (1.3) Compare the signal data obtained in step (1.2) with the nitrogen background signal data, remove the noise signal, and finally obtain the data for modeling.

4. A VOC odor rating model according to claim 3, Its characteristic is that the model During the training process, the Lasso regression method is used, that is, the L1 norm regularization term ||β|| is added. 1 The loss function is as follows: Where N represents the total number of samples used in the model training process. N samples correspond to N groups of data. n represents the nth sample, n = 1, 2, ... N, y n represents the true odor rating of the nth sample, is the odor rating predicted by the model for the nth sample, and γ is a constant coefficient.

Images