A method, system and storage medium for predicting whole vehicle VOC results

Abstract

The present application relates to a kind of method, system and storage medium for predicting whole vehicle VOC result, S1: obtaining the VOC concentration of whole vehicle and the VOC concentration of several categories of automobile parts;S2: the correlation coefficient of the VOC concentration of each category of automobile parts and the VOC concentration of whole vehicle is calculated, and the automobile parts of the category with correlation coefficient greater than set threshold value is selected;S3: algorithm model is established, the input of the algorithm model is the VOC concentration of several categories of automobile parts, and the output is the predicted value of the VOC concentration of whole vehicle, the VOC concentration of automobile parts screened in S2 is input as data set into the algorithm model and is trained, until the deviation of the predicted value of the output VOC concentration of whole vehicle of the model and the actual value of the VOC concentration of whole vehicle is in the preset range.The present application improves the risk prediction ability of whole vehicle VOC project, can quickly guide to carry out rectification action through prediction result, realizes the development efficiency improvement of whole vehicle VOC, simultaneously, it is expected to reduce the time of whole vehicle VOC test, saves whole vehicle VOC test cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of whole vehicle and parts VOC testing and evaluation, and particularly relates to a method and system for predicting whole vehicle VOC results and a storage medium. BACKGROUND

[0002] With the development of social economy, the relationship between automobiles and people is becoming closer and closer, and users' attention to automobiles has evolved from basic transportation function demand to more dimensional experience demand, and having a healthy and environmentally friendly in-vehicle space is one of the most basic and important demands. In-vehicle volatile organic compound testing (VOC) as an important evaluation index of in-vehicle health and environmental protection is widely included in the whole vehicle development evaluation index of major OEMs, and the corresponding regulations are also introduced at the national level, such as the whole vehicle standard GB / T 27630-2011 "Guidelines for the Evaluation of Air Quality in Passenger Vehicles".

[0003] For OEMs, it is necessary to have a systematic and scientific control system and method to ensure that the whole vehicle VOC level is maintained at a good level. Generally speaking, the more mature approach in the industry is to develop at the material level, the parts level, and the whole vehicle level to achieve the purpose of controlling the whole vehicle VOC level. However, this step-by-step development approach often has a long development cycle, which is difficult to support the current short cycle and fast pace of product updates for OEMs. Taking whole vehicle VOC testing as an example, considering the maturation period of various materials in the vehicle, it is necessary to carry out whole vehicle VOC testing after the state is relatively stable. The industry standard ISO 12219-12021 stipulates that the whole vehicle VOC test should be carried out 28±5 days after the vehicle is delivered. This test cycle is not friendly to short-cycle development projects, and developers cannot know the whole vehicle VOC results in a short period of time and take improvement measures.

[0004] Chinese patent CN 104111307 A discloses a method for predicting the attenuation of volatile organic compounds (VOCs) in a new vehicle. The patent establishes a VOCs emission model for automobiles by testing toluene, xylene, and ethylbenzene on vehicles delivered at 7, 10, 15, 20, 30, 45, 60, 75, and 90 days, and performing exponential fitting curve analysis on the test results. Through the emission model, the attenuation results of toluene, xylene, and ethylbenzene on vehicles delivered at any time can be predicted. However, this prediction method requires at least one whole vehicle VOC test after the parts are assembled into the whole vehicle, and cannot directly and quickly feedback the influence of parts performance on the whole vehicle VOC. At the same time, this model can only predict the emission of toluene, xylene, and ethylbenzene, and cannot predict the risk of reaching the standard of whole vehicle acetaldehyde.

[0005] Chinese patent CN 111175441 A discloses a method for pre-evaluating the air quality in a vehicle. The method determines the initial pollutant emission concentration, diffusion coefficient, separation coefficient of each vehicle interior component, and obtains the pollutant emission rate of the unit area material of each vehicle interior component by using the given mass diffusion equation of the interior of the component and the initial conditions and boundary conditions of the mass diffusion equation. Then, according to the pollutant emission rate of the unit area material of each vehicle interior component, the preset mass conservation equation and the pollutant information inside and outside the vehicle, the concentration change information of the air quality in the vehicle is obtained. This method is complex to test, requires multiple tests at multiple time points, and if the component material scheme changes, the component emission rate will change. The test sample size is large, and the actual project cannot meet the requirements, so it is not suitable for evaluating the impact of components on the whole vehicle during the project development process. SUMMARY

[0006] One of the purposes of the present application is to provide a method for predicting the VOC result of a whole vehicle to solve the problem of long cycle of obtaining VOC concentration data of the whole vehicle in the prior art. The second purpose is to provide a system for predicting the VOC result of a whole vehicle. The third purpose is to provide a storage medium.

[0007] To achieve the above purposes, the technical solutions adopted by the present application are as follows:

[0008] A method for predicting the VOC result of a whole vehicle,

[0009] S1: obtaining the VOC concentration of the whole vehicle and the VOC concentration of several categories of automobile components;

[0010] S2: calculating the correlation coefficient of the VOC concentration of each category of automobile components and the VOC concentration of the whole vehicle, and selecting the automobile components of the category whose correlation coefficient is greater than a set threshold value;

[0011] S3: establishing an algorithm model, the input of the algorithm model being the VOC concentration of several categories of automobile components, and the output being the predicted value of the VOC concentration of the whole vehicle, and inputting the VOC concentration of the automobile components selected in S2 into the algorithm model as a data set and training until the deviation between the predicted value of the VOC concentration of the whole vehicle output by the model and the actual value of the VOC concentration of the whole vehicle is within a preset range.

[0012] According to the above technical means, after obtaining the VOC concentration of the whole vehicle and the VOC concentration of several categories of automobile parts, first, the correlation coefficient is used to screen the category of the VOC concentration of the automobile parts which has a higher correlation with the VOC concentration of the whole vehicle, thereby reducing the workload of training the algorithm model. At the same time, the existing data is used to train the algorithm model until the deviation between the predicted value and the actual value is within the preset range, and then the above trained model can be used to directly obtain the VOC concentration of the whole vehicle by only obtaining the VOC concentration of the automobile parts, without the need for additional testing of the VOC concentration of the whole vehicle, thereby reducing the development cycle.

[0013] Further, the algorithm model is a regression model based on a random forest.

[0014] Compared with other machine learning models, the random forest is a model composed of the prediction results of multiple decision trees, which is superior to a single decision tree model and greatly reduces the risk of overfitting. Moreover, the random forest is very stable and has strong inclusiveness for single outliers and abnormal values in the data, and the whole model is not affected too much. Compared with the neural network model, the random forest has low calculation cost and requires less data, can deeply mine different explanations of the underlying decision tree, and can adjust the parameters based on this to have better performance. Therefore, the random forest model is used for regression prediction in the present patent.

[0015] Further, S3 specifically comprises:

[0016] S31: establishing a regression model based on a random forest;

[0017] S32: using the least mean square error for each subspace to construct a regression tree, that is, for any divided feature variable j and the corresponding split point s, the data is divided into sub-data sets R1 and R2;

[0018]

[0019] Iterate the feature j and the split point s to minimize the above formula;

[0020] In the formula, y i represents the predicted value, that is, the VOC concentration of the whole vehicle;

[0021] The predicted VOC concentration of the whole vehicle, c i represents the VOC concentration of the parts, i=1,2, and the formula (j,s) is minimized to divide the subspace, and finally n regression trees are generated; R1(j,s), R2(j,s) respectively represent different sub-data sets after division;

[0022] For each subspace, a regression tree is constructed, the random forest model combines multiple regression trees, and the final model takes the mean of all regression trees as the output;

[0023] S33: Use a portion of the dataset as the training set and the other portion as the test set;

[0024] S34: Train the random forest regression model until the deviation between the predicted value of the vehicle's VOC concentration and the actual value of the vehicle's VOC concentration is within a preset range.

[0025] Furthermore, in step S2, the correlation coefficient is obtained in the following manner;

[0026]

[0027] In the formula, PQ i The values ​​represent the VOC concentrations of automotive components, i = 1, 2, 3..., and PQ represents the VOC concentrations of the entire vehicle. This represents the correlation coefficient between the VOC concentration of automotive components and the VOC concentration of the whole vehicle.

[0028] Furthermore, in S1, all categories of automotive parts used in S2 are produced in the same batch as the complete vehicle.

[0029] Furthermore, the categories of automotive components include seats, carpets, dashboards, and headliners.

[0030] Furthermore, the preset range is -20% to 20%.

[0031] Furthermore, in S1, the VOC concentration of the whole vehicle and the VOC concentration of several categories of automotive components are derived from VOC experiments, which include:

[0032] VOC test for automotive parts: All categories of automotive parts are placed in an environment with a preset temperature. After a first preset time, the VOC concentration of each category of automotive parts is collected.

[0033] Vehicle VOC test: The whole vehicle is placed in a constant temperature and humidity environment chamber and sealed for a second preset time under preset conditions to obtain the vehicle VOC concentration.

[0034] A system for predicting the VOC result of a vehicle includes a controller and a memory, wherein the memory stores a computer-readable program that, when invoked, can execute the steps of the method for predicting the VOC result of a vehicle as described above.

[0035] A storage medium storing a computer-readable program that, when invoked, performs the steps of the method described above for predicting the VOC result of a whole vehicle.

[0036] The beneficial effects of this invention are:

[0037] 1. This invention helps improve the risk prediction capability of the whole vehicle VOC project by establishing an algorithm model and training it. The prediction results can quickly guide the implementation of rectification actions, thereby improving the efficiency of whole vehicle VOC development. At the same time, it is expected to reduce the time of whole vehicle VOC testing and save the cost of whole vehicle VOC testing.

[0038] 2. The algorithm model proposed in this invention has strong generalization ability and can be applied to different vehicle models, such as SUVs and sedans, without the need to reconstruct the model or adjust the model parameters.

[0039] 3. Given the large number of vehicle parts and the correlation between different VOC substances and parts, this invention employs feature screening before model construction. Based on the volatility characteristics of different VOC substances and the differences in part materials, correlation coefficients are used to screen parts of the vehicle that are strongly correlated with VOC substances, avoiding feature redundancy and further improving model accuracy. Attached Figure Description

[0040] Figure 1 This is a flowchart of the present invention;

[0041] Figure 2 Example of acetaldehyde prediction;

[0042] Figure 3 Example of formaldehyde prediction;

[0043] Figure 4 Example of xylene prediction;

[0044] Figure 5 Example of benzene prediction. Detailed Implementation

[0045] The following description, with reference to the accompanying drawings and preferred embodiments, illustrates the implementation of the technical solution of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0046] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0047] This embodiment proposes a method for predicting the VOC results of a vehicle, such as... Figure 1 As shown, the specific steps include:

[0048] S1: Obtain the VOC concentration of the whole vehicle and the VOC concentration of several categories of automotive parts.

[0049] This step requires collecting VOC test data for multiple complete vehicles and automotive components. For data selection, component VOC test data should be from components produced in the same batch as the vehicles used for the complete vehicle VOC test to reduce the impact of component consistency fluctuations on the results. Interior components are diverse; to ensure model accuracy, the category benchmarks for automotive components should be as comprehensive as possible during modeling, at least including components with large surface areas or volumes such as the main and auxiliary dashboards, seats, carpets, headliners, door trim panels, and sunroofs.

[0050] VOC testing method for automotive parts: Select parts produced in the same batch as those installed in the vehicle under test, and record the production date and other information of the parts in detail. Sampling is carried out on the parts after the specified curing time. The samples should be packaged in aluminum foil or PE film and then stored in an environment of (25±1)℃. The tests should be completed within 15 days according to the VOC testing enterprise standards to obtain the test data of five benzenes and three aldehydes for each part, as shown in Table 1.

[0051] Vehicle VOC testing method: After the vehicle is off the production line, it is stored in a sealed environment. After (28±5) days, the protective film on the interior surface is removed and the vehicle VOC test is carried out according to the method of HJ / T 400-2007. The test data of five benzenes and three aldehydes in the vehicle are shown in Table 1.

[0052] The VOC concentration of the whole vehicle and the VOC concentration of various types of components obtained from professional equipment testing are integrated into a dataset M{car ij C ij C ij+1 C im}, where car ij Indicates the overall VOC concentration in the vehicle, C ij The VOC concentration of the components is represented by i (i = 1, 2, ..., n), which represents different test vehicles, covering various mass-produced and under-development models such as sedans and SUVs, and j (j = 1, 2, ..., m), which represents different categories of components.

[0053] Table 1. Example of a dataset for complete vehicles and components.

[0054]

[0055]

[0056] S2: Calculate the correlation coefficient between the VOC concentration of each category of automotive parts and the VOC concentration of the whole vehicle, and select automotive parts of categories with correlation coefficients greater than a set threshold.

[0057] Because a vehicle has numerous components, and different types of automotive components produce VOCs at varying concentrations, this embodiment first performs feature screening on different components. The Pearson correlation coefficient is used to identify automotive component categories with a strong correlation to the overall vehicle's VOC levels. The specific calculation formula is as follows:

[0058]

[0059] In the formula, PQ i PQ represents the VOC concentration of automotive components, while PQ represents the VOC concentration of the entire vehicle. This represents the Pearson correlation coefficient between VOCs in automotive components and VOCs in the whole vehicle. A correlation coefficient of 1 indicates a perfect positive correlation between the two; a correlation coefficient of 0 indicates no correlation; and a correlation coefficient of -1 indicates a perfect negative correlation. In other words, the closer the correlation coefficient is to -1 or 1, the stronger the correlation between the two.

[0060] In this embodiment, a preset threshold is set for the correlation coefficient, and automotive parts with correlation coefficients greater than the preset threshold are filtered out and used as input to the model. The categories of the filtered automotive parts are shown in Table 2.

[0061] Table 2. List of VOC-related components in the vehicle

[0062] Serial number Part name 1 Instrument panel 2 Sub-instrument panel 3 Single seat 4 Three-seat 5 Door inner panel 6 Pillar 7 Trunk carpet 8 Headliner 9 Main carpet 10 Front wall panel heat insulation pad 11 Door frame sealing strip 12 Velvet groove 13 Sunroof 14 Heating, ventilation and air conditioning 15 Steering wheel

[0063] S3: Establish an algorithm model. The input of the algorithm model is the VOC concentration of several categories of automotive parts, and the output is the predicted value of the VOC concentration of the whole vehicle. The VOC concentration of automotive parts selected in S2 is used as the dataset to input the algorithm model and train it until the deviation between the predicted value of the VOC concentration of the whole vehicle and the actual value of the VOC concentration of the whole vehicle is within a preset range.

[0064] The algorithm model in this embodiment is based on a random forest regression model to predict the VOC concentration of the whole vehicle. The VOC concentrations of strongly correlated components in Table 2 are input into the model as the dataset. Random forest is a combined model that integrates multiple regression trees. Within the feature space, the model randomly divides the space into n subspaces. For each subspace, the mean squared error is used as the error factor for region segmentation, and the regression tree is constructed by minimizing the error. The method of constructing the regression tree is as follows:

[0065]

[0066] In the formula, y i c represents the predicted concentration of VOCs in the entire vehicle. i The VOC concentration of the components is represented by the formula, and the subspace is divided to minimize (j, s) in the formula, and finally n regression trees are generated.

[0067]

[0068] For each subspace, a regression tree is constructed. The random forest model combines multiple regression trees, and the final model outputs the mean of all regression trees.

[0069] The dataset is divided into training and test sets, with the training set accounting for 85% and the test set accounting for 15%. Taking formaldehyde as an example, the results of the training and test sets are presented. Using a random forest model, the VOC of the entire vehicle is predicted based on the VOC data of the components in the test set. The predicted value y is then used. pred Compared with the true value y true The comparison is performed with an allowable deviation of -20% to 20% from the true value. If the predicted value is within the range, the model is considered to be accurate; otherwise, the model is considered to be inaccurate and needs to be retrained.

[0070] S3 specifically refers to:

[0071] S31: Establish a regression model based on random forest;

[0072] S32: For each subspace, use the minimum mean square error to construct a regression tree, that is, for any feature variable j and the corresponding split point s, divide the data into subsets R1 and R2;

[0073]

[0074] Traverse the feature j and the split point s to minimize the above expression;

[0075] In the formula, y i This represents the predicted value, i.e., the VOC concentration in the entire vehicle;

[0076] Predicted VOC concentration in the whole vehicle, c i Let i = 1, 2 represent the VOC concentration of the components. The subspace is divided to minimize (j, s) in the formula, and n regression trees are generated in the end. R1(j, s) and R2(j, s) represent the different sub-data sets of the division.

[0077] For each subspace, a regression tree is constructed. The random forest model combines multiple regression trees, and the final model outputs the mean of all regression trees.

[0078] S33: Use a portion of the dataset as the training set and the other portion as the test set;

[0079] S34: Train the random forest regression model until the deviation between the predicted value of the vehicle's VOC concentration and the actual value of the vehicle's VOC concentration is within a preset range.

[0080] This embodiment constructs a correlation model between the VOC concentration of automotive parts and the VOC concentration of the whole vehicle (based on a random forest model) and uses it in actual project development. Taking some VOC substances as examples, the accuracy is 85% for formaldehyde, 80% for acetaldehyde, 75% for benzene, and 72% for xylene.

[0081] Taking aldehydes as an example ( Figure 2 and Figure 3 The acetaldehyde prediction example shows that the correlation model's predicted values ​​have a small error compared to the actual values, and some outliers are distributed around the trend line. The model's prediction accuracy for acetaldehyde is 80%. The formaldehyde prediction example shows that the correlation model underestimates some test values, i.e., the test values ​​are high, while the model's predicted values ​​are low. This is mainly due to uneven sample distribution, with a small proportion of high values. Supplementing the data sample in future work will solve this problem. Overall, the formaldehyde prediction example demonstrates high model accuracy and small prediction error.

[0082] Taking benzene compounds as an example ( Figure 4 and Figure 5 The xylene prediction example shows that some test values ​​in the correlation model are overestimated, i.e., the test values ​​are low, while the model prediction values ​​are high. This is similar to the formaldehyde example, which is a data distribution problem. This problem can be solved by increasing the data sample or adjusting the data distribution.

[0083] The method proposed in this embodiment is applicable to different vehicle models and different VOC substances, has strong generalization ability, and has the technical effect of reducing costs and increasing efficiency.

[0084] This embodiment also proposes a system for predicting the VOC result of a vehicle, including a controller and a memory. The memory stores a computer-readable program, which, when invoked, can execute the steps of the method for predicting the VOC result of a vehicle as described above.

[0085] This embodiment also proposes a storage medium storing a computer-readable program, which, when invoked, can execute the steps of the method for predicting the VOC result of the whole vehicle as described above.

[0086] The above embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention.

Claims

1. A method for predicting the VOC results of a whole vehicle, characterized in that: S1: Obtain the VOC concentration of the whole vehicle and the VOC concentration of several categories of automotive parts; S2: Calculate the correlation coefficient between the VOC concentration of each category of automotive parts and the VOC concentration of the whole vehicle, and select automotive parts of categories with correlation coefficients greater than a set threshold. S3: Establish an algorithm model. The input of the algorithm model is the VOC concentration of several categories of automotive parts, and the output is the predicted value of the VOC concentration of the whole vehicle. The VOC concentration of automotive parts selected in S2 is used as a dataset to input the algorithm model and train it until the deviation between the predicted value of the VOC concentration of the whole vehicle and the actual value of the VOC concentration of the whole vehicle is within a preset range.

2. The method for predicting the VOC result of a vehicle according to claim 1, characterized in that: The algorithm model is a regression model based on random forest.

3. The method for predicting the VOC result of a vehicle according to claim 2, characterized in that: S3 specifically refers to: S31: Establish a regression model based on random forest; S32: For each subspace, a regression tree is constructed using the minimum mean squared error. That is, for any feature variable j and its corresponding split point s, the data is divided into different subsets R1(j,s) and R2(j,s). Traverse the feature j and the split point s to minimize the above expression; In the formula, y i This represents the predicted value, i.e., the predicted concentration of VOCs in the entire vehicle. c i Let i = 1, 2, representing the VOC concentration of the components. The subspace is divided to minimize (j, s) in the formula, and finally n regression trees are generated. For each subspace, a regression tree is constructed. The random forest model combines multiple regression trees, and the final model outputs the mean of all regression trees. S33: Use a portion of the dataset as the training set and the other portion as the test set; S34: Train the random forest regression model until the deviation between the predicted value of the vehicle's VOC concentration and the actual value of the vehicle's VOC concentration is within a preset range.

4. The method for predicting the VOC result of a vehicle according to claim 1, characterized in that: In step S2, the correlation coefficient is obtained in the following manner; In the formula, PQ i The values ​​represent the VOC concentrations of automotive components, i = 1, 2, 3..., and PQ represents the VOC concentration of the entire vehicle. This represents the correlation coefficient between the VOC concentration of automotive components and the VOC concentration of the whole vehicle.

5. The method for predicting the VOC result of a vehicle according to claim 1, characterized in that: In S1, all categories of automotive parts used in S2 are produced in the same batch as the complete vehicle.

6. The method for predicting the VOC result of a vehicle according to claim 5, characterized in that: The categories of automotive components include seats, carpets, dashboards, and headliners.

7. The method for predicting the VOC result of a vehicle according to claim 1, characterized in that: The preset range is -20% to 20%.

8. The method for predicting the VOC result of a vehicle according to claim 1, characterized in that: In S1, the VOC concentration of the whole vehicle and the VOC concentration of several categories of automotive components are derived from VOC experiments, which include: VOC test for automotive parts: All categories of automotive parts are placed in an environment with a preset temperature. After a first preset time, the VOC concentration of each category of automotive parts is collected. Vehicle VOC test: The whole vehicle is placed in a constant temperature and humidity environment chamber and sealed for a second preset time under preset conditions to obtain the vehicle VOC concentration.

9. A system for predicting the VOC results of a vehicle, characterized in that: It includes a controller and a memory, the memory storing a computer-readable program that, when invoked, can perform the steps of the method for predicting the VOC result of a whole vehicle as described in any one of claims 1 to 8.

10. A storage medium, characterized in that: It contains a computer-readable program that, when invoked, performs the steps of the method for predicting the VOC result of a whole vehicle as described in any one of claims 1 to 8.

Images