A method and device for predicting molecular collision cross sections of environmentally friendly gases

Abstract

The present invention relates to the field of molecular collision technology and discloses a method and device for predicting molecular collision cross sections of environmentally friendly gases. The method obtains the electron group parameters of the gas to be determined through a pulsed Thomson experiment and extracts the characteristic vectors of the electron group parameters. A prediction model is trained and generated by obtaining known collision cross section data for the gas to be determined and combining it with collision cross section data for other gases in a preset database. After the prediction model is generated, the characteristic vectors obtained in the experiment are input into the prediction model so that the model outputs unknown collision cross section characteristic values for the gas to be determined. The unknown collision cross section data for the gas to be determined is obtained by performing data inverse processing on the unknown collision cross section characteristic values. The present invention improves the prediction accuracy of the gas collision cross section by obtaining more precise electron group parameters, thereby accelerating the acquisition of the complete collision cross section of the gas.

Description

Technical Field

[0001] The present invention relates to the field of molecular collision technology, and in particular to a method and device for predicting the molecular collision cross section of environmentally friendly gases. Background Art

[0002] The electron-molecule collision cross section (ECC) of gases is a crucial parameter in the numerical simulation of low-temperature plasmas, used for quantitative particle modeling. The collision cross section typically consists of the momentum transfer cross section for elastic scattering and the cross sections for inelastic scattering processes, including elastic collisions, vibrational excitation, electronic excitation, ionization, and adsorption. These cross sections can typically be calculated using electron beam experiments and quantum chemical theory.

[0003] Since environmentally friendly gas molecules have a large number of molecular components and atoms, both experimental methods and theoretical calculation methods take a long time, and the combined cross-section error is large, so the obtained collision cross-section of environmentally friendly gas is not very accurate. Summary of the Invention

[0004] The present invention provides a method and device for predicting the molecular collision cross section of environmentally friendly gases, which can obtain more accurate electron group parameters, improve the prediction accuracy of the gas collision cross section, and accelerate the acquisition of the complete collision cross section of the gas.

[0005] In order to solve the above technical problems, the present invention provides a method for predicting the molecular collision cross section of environmentally friendly gases, comprising:

[0006] Obtaining the electron group parameters of the gas to be determined by a pulsed Thomson experiment, and extracting a characteristic vector of the electron group parameters of the gas to be determined;

[0007] Acquire known collision cross-section data of the gas to be determined; wherein the gas to be determined includes a plurality of collision cross-sections, and the plurality of collision cross-sections include known collision cross-sections and unknown collision cross-sections;

[0008] Obtain collision cross-section data of several gases in a preset database;

[0009] Generate a prediction model using the known collision cross-section data of the gas to be determined and the collision cross-section data of the plurality of gases;

[0010] Inputting the characteristic vector of the electron group parameter of the gas to be determined into the prediction model to obtain the unknown collision cross section characteristic value of the gas to be determined;

[0011] Perform data inverse processing on the unknown collision cross-section characteristic value of the gas to be determined to obtain the unknown collision cross-section data of the gas to be determined.

[0012] Furthermore, the electron group parameters of the gas to be determined are obtained by the pulse Thomson experiment, specifically:

[0013] Obtaining a current waveform of the target gas during the molecular electron avalanche development process through a pulsed Thomson experiment; wherein the pulsed Thomson experiment mixes the target gas and a first gas at a first mixing ratio;

[0014] Fitting the current waveform to obtain the electron group parameters of the gas to be determined;

[0015] The electron group parameters include effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient.

[0016] Furthermore, the current waveform is fitted to obtain the electron group parameters of the gas to be determined, specifically:

[0017] The effective ionization rate, transit time and longitudinal electron diffusion characteristic time of the gas to be determined are obtained using the preset fitting formula. The specific formula is:

[0018]

[0019] Among them, N e (0) is the initial number of electrons; v eff is the effective ionization rate; T e is the transit time; τ D is the characteristic time of longitudinal electron diffusion; I e (t) is the electron current waveform of the electron avalanche development process measured by the pulse Thomson experiment; q0 is the electron charge; t is the current development time;

[0020] The effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient of the gas to be determined are calculated using the effective ionization rate, transit time and longitudinal electron diffusion characteristic time of the gas to be determined. The specific formula is:

[0021]

[0022]

[0023] Among them, k eff is the effective ionization rate coefficient; v eff is the effective ionization rate; N is the gas particle number density; W e is the electron drift velocity; T e is the transit time; d is the electrode distance; ND L is the density-normalized longitudinal electron diffusion coefficient; τ D is the characteristic time of longitudinal electron diffusion.

[0024] Furthermore, the prediction model is generated by utilizing the known collision cross-section data of the gas to be determined and the collision cross-section data of the plurality of gases, specifically:

[0025] generating a collision cross-section training set based on the known collision cross-section data of the gas to be determined and the collision cross-section data of the plurality of gases;

[0026] Using the preset solver, calculate the electron group parameter training set for each collision cross section in the collision cross section training set;

[0027] Training a first neural network model using the electronic group parameter training set;

[0028] After the training of the first neural network model is completed, the parameters of the first neural network model are determined to form a prediction model.

[0029] Furthermore, generating a collision cross section training set based on the known collision cross section data of the gas to be determined and the collision cross section data of the several gases is specifically as follows:

[0030] Classifying the collision cross-section data of the plurality of gases according to collision cross-section categories to obtain a plurality of groups of collision cross-section data;

[0031] Acquire first collision cross-section data of the same type as the known collision cross-section data of the gas to be determined, and perform weighted averaging processing on the first collision cross-section data in combination with the known collision cross-section data of the gas to be determined using a preset weighting function;

[0032] Using a preset weighting function, perform weighted averaging processing on each group of collision cross-section data except the first collision cross-section data;

[0033] Each group of collision cross-section data after weighted averaging is determined as the collision cross-section training set.

[0034] Furthermore, the preset solver is used to calculate the electron group parameter training set for each collision cross section in the collision cross section training set, specifically:

[0035] mixing the gas corresponding to each collision cross section with the first gas according to a first ratio;

[0036] Using the preset solver, the electron group parameters are obtained for several equally spaced and different reduced field strengths.

[0037] Furthermore, the first neural network model is trained using the electronic group parameter training set, specifically:

[0038] performing normalization processing on each electron group parameter in the electron group parameter training set;

[0039] The first neural network model is trained using the normalized electron group parameters, the training set loss function is calculated for each training process, and the weights are updated through the optimizer;

[0040] The input of the test set is set to the electron group parameters of the gas to be determined obtained through the pulse Thomson experiment, and the output is the unknown collision cross section of the gas to be determined. Since the inverse problem must satisfy the solution of the direct problem, the loss function of the test set is set to the mean square error between the electron group parameters in the electron group parameter training set and the electron group parameters of the gas to be determined;

[0041] When the training set loss function and the test set loss function no longer decrease during the training process, it is determined that the training of the first neural network model is completed.

[0042] The present invention provides a method for predicting molecular collision cross sections of environmentally friendly gases. The method obtains electron group parameters of the gas to be determined through a pulse Thomson experiment and extracts characteristic vectors of the electron group parameters. A prediction model is trained and generated by obtaining known collision cross section data of the gas to be determined and combining it with collision cross section data of other gases in a preset database. After the prediction model is generated, the characteristic vector obtained in the experiment is input into the prediction model so that the model outputs unknown collision cross section characteristic values of the gas to be determined. The unknown collision cross section data of the gas to be determined is obtained by performing data inverse processing on the unknown collision cross section characteristic values of the gas to be determined. By obtaining electron group parameters with higher precision, the prediction accuracy of the gas collision cross section is improved, thereby accelerating the acquisition of the complete collision cross section of the gas.

[0043] Accordingly, the present invention provides a device for predicting molecular collision cross sections of environmentally friendly gases, comprising: an extraction module, a first acquisition module, a second acquisition module, a model generation module, an input module, and an inverse processing module;

[0044] The extraction module is used to obtain the electron group parameters of the gas to be determined through a pulse Thomson experiment, and to extract the characteristic vector of the electron group parameters of the gas to be determined;

[0045] The first acquisition module is used to acquire known collision cross-section data of the gas to be determined; wherein the gas to be determined includes a plurality of collision cross-sections, and the plurality of collision cross-sections include known collision cross-sections and unknown collision cross-sections;

[0046] The second acquisition module is used to obtain collision cross-section data of several gases from a preset database;

[0047] The model generation module is used to generate a prediction model using the known collision cross-section data of the gas to be determined and the collision cross-section data of the multiple gases;

[0048] The input module is used to input the characteristic vector of the electron group parameter of the gas to be determined into the prediction model to obtain the unknown collision cross section characteristic value of the gas to be determined;

[0049] The inverse processing module is used to perform data inverse processing on the unknown collision cross-section characteristic value of the gas to be determined, so as to obtain the unknown collision cross-section data of the gas to be determined.

[0050] Furthermore, the extraction module includes: a waveform acquisition unit and a fitting unit;

[0051] The waveform acquisition unit is used to acquire a current waveform of the target gas during the molecular electron avalanche development process through a pulse Thomson experiment; wherein the pulse Thomson experiment mixes the target gas and the first gas at a first mixing ratio;

[0052] The fitting unit is used to fit the current waveform to obtain the electron group parameters of the gas to be determined;

[0053] The electron group parameters include effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient.

[0054] Furthermore, the model generation module includes: a generation unit, a calculation unit, a training unit and a model determination unit;

[0055] The generating unit is used to generate a collision cross section training set based on the known collision cross section data of the gas to be determined and the collision cross section data of the plurality of gases;

[0056] The calculation unit is used to calculate an electron group parameter training set for each collision cross section in the collision cross section training set using a preset solver;

[0057] The training unit is used to train the first neural network model using the electronic group parameter training set;

[0058] The model determination unit is used to determine the parameters of the first neural network model after the training of the first neural network model is completed to form a prediction model.

[0059] The present invention provides a molecular collision cross-section prediction device for environmentally friendly gases. Based on the organic combination of modules, it obtains more accurate electron group parameters, improves the prediction accuracy of gas collision cross-sections, and accelerates the acquisition of complete collision cross-sections of gases. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic flow chart of an embodiment of a method for predicting molecular collision cross sections of environmentally friendly gases provided by the present invention;

[0061] Figure 2 A schematic diagram of the collision cross-section training set provided by the present invention;

[0062] Figure 3 is a structural diagram of the first neural network model provided by the present invention;

[0063] Figure 4 A schematic diagram of a predicted collision cross section of C4F7N gas provided by the present invention;

[0064] Figure 5 This is a schematic structural diagram of an embodiment of the device for predicting molecular collision cross sections of environmentally friendly gases provided by the present invention. DETAILED DESCRIPTION

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0066] Example 1

[0067] See also Figure 1 , is a flow chart of an embodiment of a method for predicting molecular collision cross sections of environmentally friendly gases provided by the present invention. The method includes steps 101 to 106, each of which is specifically as follows:

[0068] Step 101: obtaining the electron group parameter of the gas to be determined by a pulsed Thomson experiment, and extracting the characteristic vector of the electron group parameter of the gas to be determined.

[0069] Furthermore, in the first embodiment of the present invention, the electron group parameters of the gas to be determined are obtained by a pulsed Thomson experiment, specifically:

[0070] Obtaining a current waveform of the target gas during the molecular electron avalanche development process through a pulsed Thomson experiment; wherein the pulsed Thomson experiment mixes the target gas and a first gas at a first mixing ratio;

[0071] Fitting the current waveform to obtain the electron group parameters of the gas to be determined;

[0072] The electron group parameters include effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient.

[0073] Furthermore, in the first embodiment of the present invention, the current waveform is fitted to obtain the electron group parameters of the gas to be determined, specifically:

[0074] The effective ionization rate, transit time and longitudinal electron diffusion characteristic time of the gas to be determined are obtained using the preset fitting formula. The specific formula is:

[0075]

[0076] Among them, N e (0) is the initial number of electrons; v eff is the effective ionization rate; T e is the transit time; τ D is the characteristic time of longitudinal electron diffusion; I e (t) is the electron current waveform of the electron avalanche development process measured by the pulse Thomson experiment; q0 is the electron charge; t is the current development time;

[0077] The effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient of the gas to be determined are calculated using the effective ionization rate, transit time and longitudinal electron diffusion characteristic time of the gas to be determined. The specific formula is:

[0078]

[0079] Among them, k eff is the effective ionization rate coefficient; v eff is the effective ionization rate; N is the gas particle number density; W e is the electron drift velocity; T e is the transit time; d is the electrode distance; ND L is the density-normalized longitudinal electron diffusion coefficient; τ D is the characteristic time of longitudinal electron diffusion.

[0080] As an example of the first embodiment of the present invention, taking C4F7N gas as an example, 5%, 10%, 20%, 50%, and 100% are set as mixing ratios in sequence, and C4F7N gas and Ar gas are mixed. At each mixing ratio, the current waveform of the C4F7N gas during the molecular electron avalanche development process is obtained. The effective ionization rate, transit time, and longitudinal electron diffusion characteristic time of the C4F7N gas are obtained according to the fitting formula. Then, the electron group parameters of the C4F7N gas are calculated using the calculation formula of the effective ionization rate coefficient, the electron drift velocity, and the density-normalized longitudinal electron diffusion coefficient.

[0081] Step 102: Acquire known collision cross-section data of the gas to be determined; wherein the gas to be determined includes a plurality of collision cross-sections, and the plurality of collision cross-sections include known collision cross-sections and unknown collision cross-sections.

[0082] Step 103: Acquire collision cross-section data of several gases from a preset database.

[0083] Step 104: Generate a prediction model using the known collision cross-section data of the gas to be determined and the collision cross-section data of the plurality of gases.

[0084] Furthermore, in the first embodiment of the present invention, the known collision cross-section data of the gas to be determined and the collision cross-section data of the several gases are used to generate a prediction model, specifically:

[0085] generating a collision cross-section training set based on the known collision cross-section data of the gas to be determined and the collision cross-section data of the plurality of gases;

[0086] Using the preset solver, calculate the electron group parameter training set for each collision cross section in the collision cross section training set;

[0087] Training a first neural network model using the electronic group parameter training set;

[0088] After the training of the first neural network model is completed, the parameters of the first neural network model are determined to form a prediction model.

[0089] Furthermore, in the first embodiment of the present invention, a collision cross section training set is generated based on the known collision cross section data of the gas to be determined and the collision cross section data of the several gases, specifically:

[0090] Classifying the collision cross-section data of the plurality of gases according to collision cross-section categories to obtain a plurality of groups of collision cross-section data;

[0091] Acquire first collision cross-section data of the same type as the known collision cross-section data of the gas to be determined, and perform weighted averaging processing on the first collision cross-section data in combination with the known collision cross-section data of the gas to be determined using a preset weighting function;

[0092] Using a preset weighting function, perform weighted averaging processing on each group of collision cross-section data except the first collision cross-section data;

[0093] Each group of collision cross-section data after weighted averaging is determined as the collision cross-section training set.

[0094] In the first embodiment of the present invention, various collision cross-section data of various gases are obtained in a preset database. The various collision cross-section data obtained are classified, and the collision cross-sections include elastic collision cross-sections, vibration excitation cross-sections, rotation excitation cross-sections, electronic excitation cross-sections, ionization cross-sections and adsorption cross-sections. Using a preset weighting function, weighted ensemble average processing is performed on the same cross-section data in pairs, and multiple groups of different cross-section data can be obtained by adjusting the weights. For collision cross-section data of the same category as the known collision cross-section data of the gas to be determined, a preset weighting function is used to perform a weighted geometric average on it and the known collision cross-section data of the gas to be determined, so as to limit the output cross-section to the vicinity of the known experimental results, and it is expected that the trained prediction model will limit the output in the same way. The weighting function is specifically:

[0095]

[0096] Among them, σ(ε) is the new cross-sectional data; ε is the value range of [10 -2 eV,10 3 eV]; σ1 is the cross section of gas 1; σ2 is the cross section of gas 2; ε1 is the threshold energy corresponding to cross section 1; ε2 is the threshold energy corresponding to cross section 2; is the threshold energy of the new collision cross section, r represents a free random number in the range of (0, 1).

[0097] As an example of the first embodiment of the present invention, for new gases, the currently known collision cross sections are incomplete and lack some key cross-sectional data. By weighting the existing collision cross sections with the training set, the output of the neural network can be limited, so that the predicted value is closer to the actual physical value. In this way, the uncertainty of the output can be reduced. Taking C4F7N gas as an example, assuming that the elastic collision cross section, electronic excitation cross section, 'high threshold' vibration excitation cross section, and ionization cross section of C4F7N gas are all known collision cross sections, they are weighted with a large number of the same type of collision cross sections obtained in the preset database, and the weight is set to 0.9. The large number of the same type of collision cross sections obtained in the preset database are set as the cross section of gas 1, and the elastic collision cross section, electronic excitation cross section, 'high threshold' vibration excitation cross section, and ionization cross section of C4F7N gas are set as the cross section of gas 2, and substituted into the preset weighting function for calculation. See. Figure 2 , is a schematic diagram of the collision cross section training set provided by the present invention, where Cross Section is the collision cross section; Energy is the collision cross section energy.

[0098] Furthermore, in the first embodiment of the present invention, a preset solver is used to calculate an electron group parameter training set for each collision cross section in the collision cross section training set, specifically:

[0099] mixing the gas corresponding to each collision cross section with the first gas according to a first ratio;

[0100] Using the preset solver, the electron group parameters are obtained for several equally spaced and different reduced field strengths.

[0101] As an example of the first embodiment of the present invention, according to the mixing ratio of C4F7N gas and Ar gas is 5%, 10%, 20%, 50% and 100%, the gas corresponding to each collision cross section and Ar gas are mixed according to the same mixing ratio, and the bolsig+ solver can be used to calculate the electron group parameters under 30 different reduced field strengths with equal spacing. The obtained electron group parameters include the longitudinal electron diffusion coefficient ND L , mobility μN, effective ionization rate coefficient keff Etc. Among them, the mobility μN can be obtained by the following formula:

[0102] μN=We·(E / N);

[0103] Where, μN is the mobility; We is the electron drift velocity; (E / N) is the reduced electric field strength; and E is the electric field strength.

[0104] Furthermore, in the first embodiment of the present invention, the first neural network model is trained using the electronic group parameter training set, specifically:

[0105] performing normalization processing on each electron group parameter in the electron group parameter training set;

[0106] The first neural network model is trained using the normalized electron group parameters, the training set loss function is calculated for each training process, and the weights are updated through the optimizer;

[0107] The input of the test set is set to the electron group parameters of the gas to be determined obtained through the pulse Thomson experiment, and the output is the unknown collision cross section of the gas to be determined. Since the inverse problem must satisfy the solution of the direct problem, the loss function of the test set is set to the mean square error between the electron group parameters in the electron group parameter training set and the electron group parameters of the gas to be determined;

[0108] When the training set loss function and the test set loss function no longer decrease during the training process, it is determined that the training of the first neural network model is completed.

[0109] In the first embodiment of the present invention, each electronic group parameter in the electronic group parameter training set is normalized, specifically:

[0110] For the collision cross-section training set, set the cross-section value to less than 10. -26 m 2 The data are regarded as 0, and the cross-sectional values are normalized to [-1, 1] after taking the logarithm;

[0111] For the electron transport coefficient, including mobility and density-normalized longitudinal electron diffusion coefficient, the logarithm is taken and normalized to [-1, 1];

[0112] For the effective ionization rate coefficient, k eff In [0m 3 / s,10 -25 m 3 / s] is equivalent to 10 within the range -25 m 3 / s, in [-10 -25 m 3 / s,0m 3 / s) is equivalent to -10-25 m 3 / s, and then process according to the following formula:

[0113]

[0114] Normalize k to obtain the eigenvector of the effective ionization rate coefficient.

[0115] The normalization formula is as follows:

[0116]

[0117] Among them, x is the original data of the training set, x min 、x max are the minimum and maximum values of the data with the same properties in the training set, and y is the eigenvalue after normalization.

[0118] In the first embodiment of the present invention, see Figure 3 , is a schematic diagram of the structure of the first neural network model provided by the present invention. A neural network model with a fully connected layer network as the basic architecture can be selected, with the number of hidden layers set to 3 and the number of input and output layers set to 1. An activation function is added between layers to learn the nonlinear characteristics between the data. The activation function can be selected as a swish function. Through the first neural network model, a mapping relationship f(x) = y is established, with the input x set to the electron group parameter under different reduced field strengths, and the output y to the collision cross section:

[0119]

[0120] Where (μN)1 and (μN)2 represent the mobility corresponding to different reduced field strengths; (ND L )1 and (ND L )2 represents the longitudinal electron diffusion coefficient NDL corresponding to different reduced field strengths; (k eff )1 and (k eff )2 represents the effective ionization rate coefficient corresponding to different reduced field strengths.

[0121] In the first embodiment of the present invention, the following formula can be used as the training set loss function:

[0122]

[0123] Among them, loss represents the loss function value of the training set; N represents the amount of data, y i represents the predicted cross section, σ(x i ) represents the actual cross section.

[0124] After calculating the loss function, you can choose to use the Adam optimizer to update the weights. When the loss function is less than the preset value, it is determined that the first neural network model training is completed.

[0125] In the first embodiment of the present invention, the following formula can be used as the test set loss function:

[0126]

[0127] Among them, MSE represents the loss function value of the test set; N represents the amount of data; x j represents the electronic group parameter of the gas to be determined; y j represents the electronic group parameters in the electronic group parameter training set.

[0128] Step 105: Inputting the characteristic vector of the electron group parameter of the gas to be determined into the prediction model to obtain the unknown collision cross section characteristic value of the gas to be determined.

[0129] Step 106: performing data inverse processing on the unknown collision cross-section characteristic value of the gas to be determined to obtain the unknown collision cross-section data of the gas to be determined.

[0130] In the first embodiment of the present invention, the cross-section sets obtained by experimental measurement and calculation for many gases are incomplete and inaccurate. Therefore, the method of limiting the output of the neural network by extracting the physical characteristics of the existing cross-section of the gas can effectively improve the collision cross-section set of the gas to be determined, predict unknown cross-sections and correct known cross-sections, thereby improving the certainty of the neural network prediction. After the prediction model training is completed, the characteristic vector of the gas to be determined obtained from the experiment is input into the prediction model to obtain the electron-molecule collision cross-section characteristic value of the gas to be determined. The characteristic value is subjected to data inverse processing to obtain the cross-section value predicted by the neural network. Taking C4F7N gas as an example, see Figure 4 Figure 2 is a schematic diagram of a predicted collision cross section for C₄FₐN gas provided by the present invention. Cross Section represents the collision cross section, and Energy represents the collision cross section energy. elastic CS represents the elastic collision cross section, ionization CS represents the ionization cross section, excitation CS1, excitation CS2, and excitation CS3 represent the 'low-threshold' vibrational excitation cross section, 'high-threshold' vibrational excitation cross section, and electronic excitation cross section, respectively. attachment CS1 and attachment CS2 represent the adsorption cross sections with a zero threshold and a non-zero threshold, respectively. These cross sections collectively constitute the collision cross section for C₄FₐN gas.

[0131] In summary, the first embodiment of the present invention provides a method for predicting the molecular collision cross section of an environmentally friendly gas, which obtains the electron group parameters of the gas to be determined through a pulse Thomson experiment and extracts the characteristic vectors of the electron group parameters; trains and generates a prediction model by obtaining known collision cross section data of the gas to be determined and combining it with the collision cross section data of other gases in a preset database; after the prediction model is generated, the characteristic vector obtained in the experiment is input into the prediction model so that the model outputs the unknown collision cross section characteristic value of the gas to be determined, and obtains the unknown collision cross section data of the gas to be determined by performing data inverse processing on the unknown collision cross section characteristic value of the gas to be determined; by obtaining electron group parameters with higher precision, the prediction accuracy of the gas collision cross section is improved, thereby accelerating the acquisition of the complete collision cross section of the gas.

[0132] Example 2

[0133] See also Figure 5 , is a schematic structural diagram of an embodiment of a device for predicting molecular collision cross sections of environmentally friendly gases provided by the present invention, the device comprising an extraction module 201, a first acquisition module 202, a second acquisition module 203, a model generation module 204, an input module 205, and an inverse processing module 206;

[0134] The extraction module 201 is used to obtain the electron group parameters of the gas to be determined through a pulse Thomson experiment, and extract the characteristic vector of the electron group parameters of the gas to be determined;

[0135] The first acquisition module 202 is used to acquire known collision cross-section data of the gas to be determined; wherein the gas to be determined includes a plurality of collision cross-sections, and the plurality of collision cross-sections include known collision cross-sections and unknown collision cross-sections;

[0136] The second acquisition module 203 is used to obtain collision cross-section data of several gases from a preset database;

[0137] The model generation module 204 is used to generate a prediction model using the known collision cross-section data of the gas to be determined and the collision cross-section data of the plurality of gases;

[0138] The input module 205 is used to input the characteristic vector of the electron group parameter of the gas to be determined into the prediction model to obtain the unknown collision cross section characteristic value of the gas to be determined;

[0139] The inverse processing module 206 is used to perform data inverse processing on the unknown collision cross-section characteristic value of the gas to be determined, so as to obtain the unknown collision cross-section data of the gas to be determined.

[0140] Furthermore, in the second embodiment of the present invention, the extraction module 201 includes: a waveform acquisition unit and a fitting unit;

[0141] The waveform acquisition unit is used to acquire a current waveform of the target gas during the molecular electron avalanche development process through a pulse Thomson experiment; wherein the pulse Thomson experiment mixes the target gas and the first gas at a first mixing ratio;

[0142] The fitting unit is used to fit the current waveform to obtain the electron group parameters of the gas to be determined;

[0143] The electron group parameters include effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient.

[0144] Furthermore, in a second embodiment of the present invention, the fitting unit includes: a first calculation subunit and a second calculation subunit;

[0145] The first calculation subunit is used to obtain the effective ionization rate, transit time and longitudinal electron diffusion characteristic time of the gas to be calculated using a preset fitting formula. The specific formula is:

[0146]

[0147] Among them, N e (0) is the initial number of electrons; v eff is the effective ionization rate; T e is the transit time; τ D is the characteristic time of longitudinal electron diffusion; I e (t) is the electron current waveform of the electron avalanche development process measured by the pulse Thomson experiment; q0 is the electron charge; t is the current development time;

[0148] The second calculation subunit is used to calculate the effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient of the gas to be calculated using the effective ionization rate, transit time and longitudinal electron diffusion characteristic time of the gas to be calculated. The specific formula is:

[0149]

[0150] Among them, k eff is the effective ionization rate coefficient; v eff is the effective ionization rate; N is the gas particle number density; W e is the electron drift velocity; T e is the transit time; d is the electrode distance; ND L is the density-normalized longitudinal electron diffusion coefficient; τ D is the characteristic time of longitudinal electron diffusion.

[0151] Furthermore, in the second embodiment of the present invention, the model generation module 204 includes: a generation unit, a calculation unit, a training unit and a model determination unit;

[0152] The generating unit is used to generate a collision cross section training set based on the known collision cross section data of the gas to be determined and the collision cross section data of the plurality of gases;

[0153] The calculation unit is used to calculate the electron group parameter training set for each collision cross section in the collision cross section training set using a preset solver;

[0154] The training unit is used to train the first neural network model using the electronic group parameter training set;

[0155] The model determination unit is used to determine the parameters of the first neural network model after the training of the first neural network model is completed to form a prediction model.

[0156] Furthermore, in the second embodiment of the present invention, the generating unit includes: a classification subunit, a first weighting subunit, a second weighting subunit and a determining subunit;

[0157] The classification subunit is used to classify the collision cross-section data of the plurality of gases according to the collision cross-section categories to obtain a plurality of groups of collision cross-section data;

[0158] The first weighting subunit is used to obtain first collision cross-section data of the same type as the known collision cross-section data of the gas to be determined, and perform weighted averaging processing on the first collision cross-section data in combination with the known collision cross-section data of the gas to be determined using a preset weighting function;

[0159] The second weighting subunit is used to perform weighted averaging processing on each group of collision cross-section data except the first collision cross-section data using a preset weighting function;

[0160] The determination subunit is used to determine each group of collision cross-section data after weighted averaging as a collision cross-section training set.

[0161] Furthermore, in the second embodiment of the present invention, the calculation unit includes: a mixing subunit and an extraction subunit;

[0162] The mixing subunit is configured to mix the gas corresponding to each collision cross section with the first gas according to a first ratio;

[0163] The extraction subunit is used to obtain the electron group parameters under several equally spaced and different reduced field strengths using a preset solver.

[0164] Furthermore, in the second embodiment of the present invention, the training unit includes: a normalization subunit, a training subunit, a testing subunit and a judgment subunit;

[0165] The normalization subunit is used to perform normalization processing on each electronic group parameter in the electronic group parameter training set;

[0166] The training subunit is used to train the first neural network model using the normalized electronic group parameters, calculate the training set loss function of each training process, and update the weights through the optimizer;

[0167] The test subunit is used to set the input of the test set to the electron group parameters of the gas to be determined obtained through the pulse Thomson experiment, and the output to be the unknown collision cross section of the gas to be determined. According to the inverse problem must satisfy the solution of the direct problem, the test set loss function is set to the mean square error between the electron group parameters in the electron group parameter training set and the electron group parameters of the gas to be determined;

[0168] The judgment subunit is used to determine that the training of the first neural network model is completed when the training set loss function and the test set loss function no longer decrease during the training process.

[0169] In summary, the second embodiment of the present invention provides a molecular collision cross-section prediction device for environmentally friendly gases. Based on the organic combination of modules, the electron group parameters of the gas to be determined are obtained through a pulse Thomson experiment, and the characteristic vectors of the electron group parameters are extracted; by obtaining the known collision cross-section data of the gas to be determined and combining it with the collision cross-section data of other gases in a preset database, a prediction model is trained and generated; after the prediction model is generated, the characteristic vector obtained in the experiment is input into the prediction model so that the model outputs the unknown collision cross-section characteristic value of the gas to be determined, and the unknown collision cross-section data of the gas to be determined are obtained by performing data inverse processing on the unknown collision cross-section characteristic value of the gas to be determined; by obtaining electron group parameters with higher precision, the prediction accuracy of the gas collision cross-section is improved, thereby accelerating the acquisition of the complete collision cross-section of the gas.

[0170] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for predicting molecular collision cross sections of environmentally friendly gases, characterized in that: include: Obtaining the electron group parameters of the gas to be determined by a pulsed Thomson experiment, and extracting a characteristic vector of the electron group parameters of the gas to be determined; Acquire known collision cross-section data of the gas to be determined; wherein the gas to be determined includes a plurality of collision cross-sections, and the plurality of collision cross-sections include known collision cross-sections and unknown collision cross-sections; Obtain collision cross-section data of several gases in a preset database; Generate a prediction model using the known collision cross-section data of the gas to be determined and the collision cross-section data of the plurality of gases; Inputting the characteristic vector of the electron group parameter of the gas to be determined into the prediction model to obtain the unknown collision cross section characteristic value of the gas to be determined; Performing data inverse processing on the unknown collision cross-section characteristic value of the gas to be determined to obtain unknown collision cross-section data of the gas to be determined; The electron group parameters of the gas to be determined are obtained by pulse Thomson experiment, specifically: Obtaining a current waveform of the target gas during the molecular electron avalanche development process through a pulsed Thomson experiment; wherein the pulsed Thomson experiment mixes the target gas and a first gas at a first mixing ratio; Fitting the current waveform to obtain the electron group parameters of the gas to be determined; The electron group parameters include effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient.

2. The method for predicting molecular collision cross sections of environmentally friendly gases according to claim 1, characterized in that: The current waveform is fitted to obtain the electron group parameters of the gas to be determined, specifically: The effective ionization rate, transit time and longitudinal electron diffusion characteristic time of the gas to be determined are obtained using the preset fitting formula. The specific formula is: Among them, N e (0) is the initial number of electrons; v eff is the effective ionization rate; T e is the transit time; τ D is the characteristic time of longitudinal electron diffusion; I e (t) is the electron current waveform of the electron avalanche development process measured by the pulse Thomson experiment; q0 is the electron charge; t is the current development time; The effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient of the gas to be determined are calculated using the effective ionization rate, transit time and longitudinal electron diffusion characteristic time of the gas to be determined. The specific formula is: Among them, k eff is the effective ionization rate coefficient; v eff is the effective ionization rate; N is the gas particle number density; W e is the electron drift velocity; T e is the transit time; d is the electrode distance; ND L is the density-normalized longitudinal electron diffusion coefficient; τ D is the characteristic time of longitudinal electron diffusion.

3. The method for predicting molecular collision cross sections of environmentally friendly gases according to claim 1, characterized in that: The method of generating a prediction model by using the known collision cross-section data of the gas to be determined and the collision cross-section data of the plurality of gases is as follows: generating a collision cross-section training set based on the known collision cross-section data of the gas to be determined and the collision cross-section data of the plurality of gases; Using the preset solver, calculate the electron group parameter training set for each collision cross section in the collision cross section training set; Training a first neural network model using the electronic group parameter training set; After the training of the first neural network model is completed, the parameters of the first neural network model are determined to form a prediction model.

4. The method for predicting molecular collision cross sections of environmentally friendly gases according to claim 3, characterized in that: The generation of a collision cross section training set based on the known collision cross section data of the gas to be determined and the collision cross section data of the several gases is specifically as follows: Classifying the collision cross-section data of the plurality of gases according to collision cross-section categories to obtain a plurality of groups of collision cross-section data; Acquire first collision cross-section data of the same type as the known collision cross-section data of the gas to be determined, and perform weighted averaging processing on the first collision cross-section data in combination with the known collision cross-section data of the gas to be determined using a preset weighting function; Using a preset weighting function, perform weighted averaging processing on each group of collision cross-section data except the first collision cross-section data; Each group of collision cross-section data after weighted averaging is determined as the collision cross-section training set.

5. The method for predicting molecular collision cross sections of environmentally friendly gases according to claim 3, characterized in that: The preset solver is used to calculate the electron group parameter training set for each collision cross section in the collision cross section training set, specifically: mixing the gas corresponding to each collision cross section with the first gas according to a first ratio; Using the preset solver, the electron group parameters are obtained for several equally spaced and different reduced field strengths.

6. The method for predicting molecular collision cross sections of environmentally friendly gases according to claim 3, characterized in that: The first neural network model is trained using the electronic group parameter training set, specifically: performing normalization processing on each electron group parameter in the electron group parameter training set; The first neural network model is trained using the normalized electron group parameters, the training set loss function is calculated for each training process, and the weights are updated through the optimizer; The input of the test set is set to the electron group parameters of the gas to be determined obtained through the pulse Thomson experiment, and the output is the unknown collision cross section of the gas to be determined. Since the inverse problem must satisfy the solution of the direct problem, the loss function of the test set is set to the mean square error between the electron group parameters in the electron group parameter training set and the electron group parameters of the gas to be determined; When the training set loss function and the test set loss function no longer decrease during the training process, it is determined that the training of the first neural network model is completed.

7. A molecular collision cross section prediction device for environmentally friendly gases, characterized in that: include: extraction module, first acquisition module, second acquisition module, model generation module, input module and inverse processing module; The extraction module is used to obtain the electron group parameters of the gas to be determined through a pulse Thomson experiment, and to extract the characteristic vector of the electron group parameters of the gas to be determined; The first acquisition module is used to acquire known collision cross-section data of the gas to be determined; wherein the gas to be determined includes a plurality of collision cross-sections, and the plurality of collision cross-sections include known collision cross-sections and unknown collision cross-sections; The second acquisition module is used to obtain collision cross-section data of several gases from a preset database; The model generation module is used to generate a prediction model using the known collision cross-section data of the gas to be determined and the collision cross-section data of the multiple gases; The input module is used to input the characteristic vector of the electron group parameter of the gas to be determined into the prediction model to obtain the unknown collision cross section characteristic value of the gas to be determined; The inverse processing module is used to perform data inverse processing on the unknown collision cross-section characteristic value of the gas to be determined, so as to obtain the unknown collision cross-section data of the gas to be determined; Wherein, the extraction module includes: a waveform acquisition unit and a fitting unit; The waveform acquisition unit is used to acquire a current waveform of the target gas during the molecular electron avalanche development process through a pulse Thomson experiment; wherein the pulse Thomson experiment mixes the target gas and the first gas at a first mixing ratio; The fitting unit is used to fit the current waveform to obtain the electron group parameters of the gas to be determined; The electron group parameters include effective ionization rate coefficient, electron drift velocity and density-normalized longitudinal electron diffusion coefficient.

8. The device for predicting molecular collision cross sections of environmentally friendly gases according to claim 7, characterized in that: The model generation module includes: a generation unit, a calculation unit, a training unit and a model determination unit; The generating unit is used to generate a collision cross section training set based on the known collision cross section data of the gas to be determined and the collision cross section data of the plurality of gases; The calculation unit is used to calculate an electron group parameter training set for each collision cross section in the collision cross section training set using a preset solver; The training unit is used to train the first neural network model using the electronic group parameter training set; The model determination unit is used to determine the parameters of the first neural network model after the training of the first neural network model is completed to form a prediction model.

Images