A method for predicting the oxidation reaction activity of particulate matter

Abstract

The present invention relates to the technical field of particle characteristic analysis, and discloses a method for predicting the oxidation reaction activity of particles. Through the RCCI engine test platform, a particle collection device is used to collect particle samples under different premixing ratios, and the particle physical characteristic parameters (crystal length, crystallite curvature, crystallite interlayer spacing) and oxidation reaction activity parameters are collected to establish a data set, and then multiple sub-training sets are extracted from the data set, and the data in each sub-training set is input into the corresponding base learner, and an equation with the same multivariate relationship is defined in each base learner as a weak predictor to predict the oxidation reaction activity of the particles. The present invention can predict the oxidation reaction activity of the particles based only on their physical characteristics, greatly reducing the experimental time, saving the experimental cost, and improving the experimental efficiency.

Description

Technical Field

[0001] The invention relates to the technical field of particle property analysis, and in particular to a method for predicting the oxidation reaction activity of particles. Background Art

[0002] In the construction of global urban industrialization and modernization, particulate matter is one of the sources of air pollution. It has the characteristics of small particle size, containing a large amount of toxic substances, long atmospheric residence time, and long transportation distance, which poses a huge threat to the quality of the atmospheric environment, especially human health. RCCI (Reactivity Controlled Compression Ignition) is an efficient and environmentally friendly combustion strategy. In the face of the current lack of fossil energy and increasingly serious environmental pollution, the low-temperature premixed combustion mode adopted by automobiles has the characteristics of high thermal efficiency and the ability to control the combustion phase, which can effectively reduce the generation of exhaust particulate matter.

[0003] In the current study of the physical and chemical properties of RCCI exhaust particulate matter, TEM (Transmission Electron Microscope) is generally used to analyze the microscopic morphology, basic carbon particle diameter, nanostructure and other physical parameters of the particulate matter. This method has a short sampling time of 10-15S; while the determination of the oxidation reaction activity parameters requires the extraction of particulate matter with a special filter membrane. The sampling time under this method is longer than the former, about 15min. After the sample is dried after collection and the impurities in the sample are removed, the oxidation reaction activity of the particulate matter needs to be analyzed by a thermogravimetric analyzer. Therefore, from sampling to analysis, the time to obtain the oxidation reaction activity parameters of the particulate matter is much longer than the time to obtain the physical property parameters of the particulate matter. The entire particulate matter experimental cycle is long, and the time cost invested is high. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a method for predicting the oxidation reaction activity of particulate matter, establish a database with the existing physical property parameters of particulate matter (crystallite length, crystallite curvature, crystallite interlayer spacing) and the data of oxidation reaction activity parameters, and use the bagging algorithm to find the relationship between the parameters to achieve the purpose of predicting the oxidation reaction activity of particulate matter and improve the experimental efficiency. The technical solution is as follows:

[0005] A method for predicting the oxidation reaction activity of particulate matter comprises the following steps:

[0006] Step 1: Using the RCCI engine test platform, a particle collection device is used to collect particle samples for analyzing the physical characteristics of particles and particle samples for analyzing the oxidation reaction activity of particles at different premixing ratios;

[0007] Step 2: Obtain data on crystallite length, crystallite curvature and crystallite interlayer distance from the particle samples used to analyze the physical characteristics of the particles; obtain data on oxidation reaction activity from the particle samples used to analyze the oxidation reaction activity of the particles, and establish a data set;

[0008] Step 3: extracting data of each category from the data set multiple times with replacement, and taking the data extracted each time as a sub-training set, and ensuring that each sub-training set includes 16 data of four categories, namely, oxidation reaction activity, crystallite length, crystallite interlayer spacing and crystallite curvature, and extracting N times to obtain N sub-training sets;

[0009] Step 4: Input the data in each sub-training set into the corresponding base learner, and define an equation of the same multivariate relationship as a weak predictor in each base learner;

[0010] The oxidation reaction activity parameter is used as the dependent variable Y, and the crystallite length A, crystallite curvature B and crystallite interlayer distance C are used as independent variables. Each independent variable is multiplied by an unknown coefficient θ1, θ2 and θ3, and an error term W0 is introduced. The relationship is expressed as follows:

[0011] ;

[0012] In each base learner, a method capable of solving the respective error terms W is established according to the 16 data. 0,i , coefficient θ 1,i ,θ 2,i and θ 3,i The four-element system of equations, i=1,2…N, represents the i-th base learner;

[0013] The error term W obtained from solving each base learner 0,i , coefficient θ 1,i ,θ 2,i and θ 3,i Take the average value to get the average value of the error term , and the coefficient average , and ;

[0014] Step 5: Obtain the relationship between the oxidation reaction activity of the reaction particles and the crystallite length, crystallite curvature and crystallite interlayer spacing:

[0015] ;

[0016] This formula can be used to predict the oxidation reaction activity of particles when the physical characteristics of the particles, namely, crystallite length A, crystallite curvature B and crystallite interlayer spacing C, are known.

[0017] Furthermore, the step 1 specifically includes:

[0018] Step 1.1: Collect particle samples for analysis of particle physical characteristics:

[0019] Use tape to stick the dried TEM support film onto filter paper and fix it in the particle collection device; use a vacuum pump to directly extract the engine exhaust gas and let the exhaust gas flow through the particle collection device to obtain samples of particles under different working conditions;

[0020] Step 1.2: Collect PM samples for analysis of PM oxidation activity:

[0021] The engine exhaust is extracted by a vacuum pump and then passed through the filter membrane to adhere the particles to the filter membrane. The filter membrane is then placed in a constant temperature drying oven to dry, and the particles are then separated from the dried filter membrane.

[0022] Furthermore, the separation of particles from the dried filter membrane in step 1.2 is specifically as follows:

[0023] Step 1.2.1: The dried filter membrane is placed in an ultrasonic machine containing dichloromethane for 1 hour to separate the particles on the filter membrane through ultrasonic vibration, and at the same time, the SOF in the particles is dissolved in the dichloromethane;

[0024] Step 1.2.2: Prepare the dichloromethane solution in a clean glass centrifuge tube and centrifuge it for 2 hours to separate the solid particles from the solution, so that the particles adhere to the centrifuge tube; remove the solution above to obtain clean soot particles without SOF;

[0025] Step 1.2.3: Repeat steps 1.2.1 and 1.2.2 at least 3 times.

[0026] Furthermore, the step 2 is specifically as follows:

[0027] Step 2.1: Obtain physical characteristic data of particles:

[0028] The TEM images of the particles were obtained by low-magnification electron microscopy, and the complete microscopic morphology of the particles was obtained at the micropores of the support membrane;

[0029] The HRTEM images of the particles are obtained by high-power electron microscopy, and the microscopic morphology of the basic carbon particles with complete structure is obtained at the edge of the particles;

[0030] The data of crystallite length, crystallite curvature and crystallite interlayer distance of the particle sample were extracted by MicrocrystallineImgProcess4.0;

[0031] Step 2.2: Obtain oxidation activity data of particles:

[0032] The oxidation reaction activity of the particulate matter is obtained by a thermogravimetric analyzer. According to the relationship between the mass loss of soot particles and the temperature, the oxidation reaction activity of the particulate matter is quantitatively characterized by analyzing the function curve of the mass change with temperature.

[0033] Furthermore, in step 4, the first base learner solves the following set of equations to obtain the first set of error terms W 0,1 and coefficient θ 1,1 ,θ 2,1 and θ 3,1 The value of

[0034] ;

[0035] Among them, A 1,1 , A 2,1 , A 3,1 and A 4,1 are the four data of crystallite length in the first sub-training set; B 1,1 , B 2,1 , B 3,1 and B 4,1 are the four data of crystallite curvature in the first sub-training set; C 1,1 , C 2,1 , C 3,1 and C 4,1 are the four data of the microcrystalline layer spacing in the first sub-training set; Y 1,1 , Y 2,1 , Y 3,1 and Y 4,1 are the four data of oxidation reaction activity in the first sub-training set; the first digit of the subscript of the above data indicates the data number, and the second digit of the subscript indicates the base learner number; the coefficient θ 1,1 ,θ 2,1 ,θ 3,1 and the error term W 0,1 The second digit of the subscript in represents the base learner number;

[0036] Similarly, the i-th base learner solves the following set of equations to obtain the i-th group of error terms W 0,i and coefficient θ 1,i ,θ 2,i and θ 3,i The value of

[0037] ;

[0038] Under N base learners, we get N pairs of W 0,i ,θ 1,i ,θ 2,i ,θ 3,i The value of , the average value is as follows:

[0039] ;

[0040] Get the mean error term , and the coefficient average , and .

[0041] Compared with the prior art, the beneficial effect of the present method is as follows: the present invention establishes a database based on the parameters characterizing the physical properties and oxidation reaction activity of the particulate matter collected under different premixing ratios by a machine learning method, and obtains a relatively stable relationship between the oxidation reaction activity of the particulate matter and the crystallite length, crystallite curvature, and crystallite interlayer spacing through a bagging algorithm, thereby reducing the experimental time for analyzing the oxidation reaction activity of the particulate matter, and being able to predict the oxidation reaction activity of the particulate matter based solely on its physical characteristics in subsequent experiments, thereby greatly reducing the experimental time, saving experimental costs, and improving experimental efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The present invention is a flow chart of the method for predicting the oxidation reaction activity of particulate matter. DETAILED DESCRIPTION

[0043] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. The present invention establishes a database of the physical properties of particulate matter and the oxidation reaction activity, and implements a flow chart for predicting the oxidation reaction activity under machine learning. Figure 1 As shown in the flow chart, it is necessary to first collect particulate matter for extracting particulate matter physical characteristic parameters and oxidation reaction activity parameters under the RCCI engine experimental platform, and then use special instruments (field emission transmission electron microscope, thermogravimetric analyzer) to obtain the required parameters of crystallite length, crystallite curvature, interlayer spacing, and oxidation reaction activity quantitative data to establish a database.

[0044] 1. Establish a database

[0045] The specific technical solutions adopted are:

[0046] 1. Through the RCCI engine experimental platform, the particulate matter under different RP (premixing ratio) was collected with a homemade particulate matter collection device.

[0047] First, the particle samples for analyzing the physical characteristics of the particles were collected, and the dried TEM support film was adhered to the filter paper using tape and fixed in the homemade collection device.

[0048] The engine exhaust is directly extracted by a vacuum pump and made to flow through the particulate matter collection device to obtain particulate matter samples under different working conditions (the low-activity fuel RP is 30%, 40% and 50% respectively). The engine exhaust is controlled and the flow sampling time is controlled within 8~10S.

[0049] The TEM image of the particles was obtained by low-magnification electron microscopy, and the complete microscopic morphology of the particles was obtained at the micropores of the TEM support membrane (magnification was 10,000 to 100,000 times).

[0050] The HRTEM images of the particles were obtained by high-power electron microscopy (magnification of 200,000 to 400,000 times), and the microscopic morphology of the basic carbon particles with less overlap and more complete structure was obtained at the edges of the particles.

[0051] The data of crystallite length, crystallite curvature and crystallite interlayer spacing of the particle samples were extracted using MicrocrystallineImgProcess4.0.

[0052] The same collector is used to collect particulate samples for analyzing the oxidation reaction activity of particulate matter. A special filter membrane is made of tetrafluoroethylene (chemical formula C2F4), which is inert to most chemicals and solvents, resistant to strong acids and alkalis, water and organic solvents, and has high chemical stability; it has excellent corrosion resistance and heat resistance (the operating temperature can reach 250°C). The filter membrane using this material facilitates the subsequent extraction of particulate matter.

[0053] The engine exhaust is directly extracted by a vacuum pump, and the exhaust flows through the filter membrane, causing the particles to adhere to the filter membrane. The collection time is about 30 minutes, and then the filter membrane is placed in a constant temperature drying oven at 60°C for 30 minutes to remove moisture from the exhaust.

[0054] After drying, the filter membrane was placed in an ultrasonic machine containing dichloromethane for ultrasonic vibration for 1 hour, and the particles on the filter membrane were separated by ultrasonic vibration.

[0055] At the same time, the SOF (Soluble Organic Fractions) in the particles was dissolved in dichloromethane. The dichloromethane solution was prepared in a 10ml clean glass centrifuge tube and centrifuged for 30min.

[0056] Due to the centrifugal force, the solid particles and the solution have been separated, and the particles are firmly attached to the centrifuge tube. After removing the solution above, what remains is clean soot particles without SOF.

[0057] To ensure that the SOF on the particles was completely removed, the water bath sonication and centrifugation analysis steps were repeated at least three times.

[0058] Finally, the clean, SOF-free particles were placed in a refrigerator for testing.

[0059] The oxidation reaction activity of particulate matter is obtained through a thermogravimetric analyzer, and the relationship between the mass loss of soot particles and temperature is explored. By analyzing the function curve of the mass change with temperature, the oxidation reaction activity of the particulate matter can be quantitatively characterized.

[0060] A database is established for use based on the data of four parameters of the particles, namely, crystallite length, crystallite curvature, crystallite layer spacing, and oxidation reaction activity, collected under certain working conditions.

[0061] 2. Obtaining prediction results through machine learning

[0062] The Bagging algorithm (Bootstrap Aggregating), also known as the bagging algorithm, is a parallel learning algorithm in the field of machine learning. Its main idea is to repeatedly sample in samples, randomly extract multiple times with replacement from the original sample data set, and use the extracted data as training sample subsets. The number of training sample subsets is the same as the number of extractions. These training sample subsets are then used to train base learners, and multiple base learners are used to jointly predict the results.

[0063] Extract the same type and number of data from the dataset multiple times with replacement. The data extracted each time is used as a sub-training set. The number of sub-training sets should be as large as possible.

[0064] The data in each sub-training set is input into the corresponding base learner, and an equation of the same multivariate relationship is defined in each base learner as a weak predictor to prepare for the subsequent exploration of the relationship between the parameters.

[0065] Each base learner sub-training set input can output a result based on the data with four parameters.

[0066] Since the extraction is done with replacement, there may be a situation where the data extracted from a sub-training set is the same as that from another sub-training set. Therefore, in order to reduce the error, the number of sub-training sets should be large enough.

[0067] Ideally, the data extracted from each sub-training set will contain four data of each of the four parameters, which will be input into the base learner, and the values ​​of the coefficients and error terms of each parameter in the sub-training set can be obtained.

[0068] Specifically, extract four parameters of each data N times with replacement in the data set to ensure that each sub-training set contains data of four categories: oxidation reaction activity, crystallite length, crystallite interlayer spacing, and crystallite curvature. Take the 16 data with four parameters extracted each time as a sub-training set, and extract N sub-training sets N times. Since the data samples in the sub-training are randomly extracted, there is a situation where part or all of the extracted data is the same as the data in other sub-training sets. In order to reduce the impact of this situation on the prediction results, the number of sub-training sets N should be large.

[0069] Since the prediction effect of a base learner is based on less data, there are certain differences in the results predicted by each base learner.

[0070] In order to make the prediction results stable, when enough prediction results of the base learner are output, most of the prediction results will be concentrated near the actual numerical values ​​of the oxidation reaction activity of the particles and the crystallite length, crystallite curvature, and crystallite interlayer spacing.

[0071] Since each sub-training set has parameters for all types of input base learners, the oxidation reaction activity parameters of the data input into the base learner of each sub-training set are used as the dependent variable Y, and the crystallite length A, crystallite curvature B and crystallite interlayer spacing C are used as independent variables. Accordingly, each independent variable should be multiplied by an unknown coefficient θ1, θ2 and θ3, and then an error term W0 is introduced. The relationship is expressed as:

[0072] ;

[0073] Using the input data, four solvable multivariate relationships between oxidation reaction activity and crystallite length, crystallite curvature, and crystallite interlayer spacing can be established in the base learner.

[0074] The error term W can be obtained in the first base learner 0,1 and coefficient θ 1,1 ,θ 2,1 and θ 3,1 The values ​​of the four parameters are:

[0075] ;

[0076] Among them, A 1,1 , A 2,1 , A 3,1 and A 4,1 are the four data of crystallite length in the first sub-training set; B 1,1 , B 2,1 , B 3,1 and B 4,1 are the four data of crystallite curvature in the first sub-training set; C 1,1 , C2,1 , C 3,1 and C 4,1 are the four data of the microcrystalline layer spacing in the first sub-training set; Y 1,1 , Y 2,1 , Y 3,1 and Y 4,1 These are the four data of oxidation reaction activity in the first sub-training set.

[0077] Note: The first digit of the above data subscript indicates the data sequence number, N sub-training sets, a total of N+3 groups of data, and the second digit indicates which base learner it belongs to. The coefficient θ 1,1 ,θ 2,1 ,θ 3,1 and the error term W 0,1 The second digit of the subscript in indicates which base learner it belongs to.

[0078] Similarly, the i-th base learner solves the following set of equations to obtain the i-th group of error terms W 0,i and coefficient θ 1,i ,θ 2,i and θ 3,i The value of

[0079] ;

[0080] Correspondingly, the Nth base learner can obtain W in the data 0,N ,θ 1,N ,θ 2,N ,θ 3,N The value of , under N base learners, N pairs of W can be obtained respectively 0,i ,θ 1,i ,θ 2,i ,θ 3,i The value of:

[0081] ;

[0082] Get the mean error term , and the coefficient average , and .

[0083] Finally, we have: .

[0084] Based on the method of the present invention, most of the data in the training set are taken into consideration, and finally the relationship between the oxidation reaction activity of the reactive particles and the crystallite length, crystallite curvature and crystallite interlayer spacing can be obtained, which can be used as a method for predicting the oxidation reaction activity of the particles when the physical characteristics of the particles (crystallite length, crystallite curvature, crystallite interlayer spacing) are known.

Claims

1. A method for predicting the oxidation reaction activity of particulate matter, characterized in that: The following steps are involved: Step 1: Using the RCCI engine test platform, a particle collection device is used to collect particle samples for analyzing the physical characteristics of particles and particle samples for analyzing the oxidation reaction activity of particles at different premixing ratios; Step 2: Obtain data on crystallite length, crystallite curvature and crystallite interlayer distance from the particle samples used to analyze the physical characteristics of the particles; obtain data on oxidation reaction activity from the particle samples used to analyze the oxidation reaction activity of the particles, and establish a data set; The specific data of the oxidation reaction activity of the particulate matter are as follows: The oxidation reaction activity of the particulate matter is obtained by a thermogravimetric analyzer. Based on the relationship between the mass loss of soot particles and the temperature, the oxidation reaction activity of the particulate matter is quantitatively characterized by analyzing the function curve of the mass change with the temperature. Step 3: extracting data of each category from the data set multiple times with replacement, and taking the data extracted each time as a sub-training set, and ensuring that each sub-training set includes 16 data of four categories, namely, oxidation reaction activity, crystallite length, crystallite interlayer spacing and crystallite curvature, and extracting N times to obtain N sub-training sets; Step 4: Input the data in each sub-training set into the corresponding base learner, and define an equation of the same multivariate relationship as a weak predictor in each base learner; The oxidation reaction activity parameter is used as the dependent variable Y, and the crystallite length A, crystallite curvature B and crystallite interlayer distance C are used as independent variables. Each independent variable is multiplied by an unknown coefficient θ1, θ2 and θ3, and an error term W0 is introduced. The relationship is expressed as follows: θ1*A+θ2*B+θ3*C+W0=Y In each base learner, a method capable of solving the respective error terms W is established according to the 16 data. 0,i , coefficient θ 1i ,θ 2i and θ 3i The four-element system of equations, i = 1, 2…N, represents the i-th basis learner; The error term W obtained from solving each base learner 0,i , coefficient θ 1,i ,θ 2,i and θ 3,i Take the average value to get the average value of the error term And the coefficient average and Step 5: Obtain the relationship between the oxidation reaction activity of the reaction particles and the crystallite length, crystallite curvature and crystallite interlayer spacing: This formula can be used to predict the oxidation reaction activity of particles when the physical characteristics of the particles, namely, crystallite length A, crystallite curvature B and crystallite interlayer spacing C, are known.

2. The method for predicting the oxidation reaction activity of particulate matter according to claim 1, characterized in that: The step 1 specifically includes: Step 1.1: Collect particle samples for analysis of particle physical characteristics: The dried TEM support film is attached to the filter paper using adhesive tape and fixed in the particle collection device; the engine exhaust is directly extracted using a vacuum pump and the exhaust gas is passed through the particle collection device to obtain samples of particles under different working conditions; Step 1.2: Collect PM samples for analysis of PM oxidation activity: The engine exhaust is extracted by a vacuum pump and then passed through the filter membrane to adhere the particles to the filter membrane. The filter membrane is then placed in a constant temperature drying oven to dry, and the particles are then separated from the dried filter membrane.

3. The method for predicting the oxidation reaction activity of particulate matter according to claim 2, characterized in that: The separation of particles from the dried filter membrane in step 1.2 is specifically as follows: Step 1.2.1: The dried filter membrane is placed in an ultrasonic machine containing dichloromethane for 1 hour to separate the particles on the filter membrane through ultrasonic vibration, and at the same time, the SOF in the particles is dissolved in the dichloromethane; Step 1.2.2: Prepare the dichloromethane solution into a clean glass centrifuge tube, centrifuge it for 2 hours to separate the solid particles from the solution, so that the particles adhere to the centrifuge tube; remove the solution above to obtain clean soot particles without SOF; Step 1.2.3: Repeat step 1.2.1 and step 1.2.2 at least 3 times.

4. The method for predicting the oxidation reaction activity of particulate matter according to claim 1, characterized in that: The physical characteristic data of the particles obtained in step 2 are specifically: The TEM image of the particles was obtained by low-magnification electron microscopy, and the complete microscopic morphology of the particles was obtained at the micropores of the support membrane; The HRTEM images of the particles are obtained by high-power electron microscopy, and the microscopic morphology of the basic carbon particles with complete structure is obtained at the edge of the particles; The data of crystallite length, crystallite curvature and crystallite interlayer spacing of the particle samples were extracted using MicrocrystallineImgProcess4.

0.

5. The method for predicting the oxidation reaction activity of particulate matter according to claim 1, characterized in that: In step 4, the first base learner solves the following set of equations to obtain the first set of error terms W 0,1 and coefficient θ 11 ,θ 21 and θ 31 The value of Among them, A 1,1 , A 2,1 , A 3,1 and A 4,1 are the four data of crystallite length in the first sub-training set; B 1,1 , B 2,1 , B 3,1 and B 4,1 are the four data of crystallite curvature in the first sub-training set; C 1,1 , C 2,1 , C 3,1 and C 4,1 are the four data of microcrystalline layer spacing in the first sub-training set; Y 1,1 , Y 2,1 , Y 3,1 and Y 4,1 are the four data of oxidation reaction activity in the first sub-training set; the first digit of the subscript of the above data indicates the data number, and the second digit of the subscript indicates the base learner number; the coefficient θ 1,1 ,θ 2,1 ,θ 3,1 and the error term W 0,1 The second digit of the subscript in represents the base learner number; Similarly, the i-th base learner solves the following set of equations to obtain the i-th group of error terms W 0,i and coefficient θ 1,i ,θ 2,i and θ 3,i The value of Under N base learners, we get N pairs of W 0,i ,θ 1,i ,θ 2,i ,θ 3,i The value of , the average value is as follows: Get the mean error term And the coefficient average and

Images