CN111755080A - Method for predicting adsorption performance of MOF (metal organic framework) on methane gas based on deep convolutional neural network - Google Patents

Method for predicting adsorption performance of MOF (metal organic framework) on methane gas based on deep convolutional neural network Download PDF

Info

Publication number
CN111755080A
CN111755080A CN202010374618.XA CN202010374618A CN111755080A CN 111755080 A CN111755080 A CN 111755080A CN 202010374618 A CN202010374618 A CN 202010374618A CN 111755080 A CN111755080 A CN 111755080A
Authority
CN
China
Prior art keywords
mof
classifier
neural network
convolutional neural
methane gas
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010374618.XA
Other languages
Chinese (zh)
Other versions
CN111755080B (en
Inventor
卢罡
赵正阳
阳庆元
李睿琪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Chemical Technology
Original Assignee
Beijing University of Chemical Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing University of Chemical Technology filed Critical Beijing University of Chemical Technology
Priority to CN202010374618.XA priority Critical patent/CN111755080B/en
Publication of CN111755080A publication Critical patent/CN111755080A/en
Application granted granted Critical
Publication of CN111755080B publication Critical patent/CN111755080B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C60/00—Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
    • G01N15/08—Investigating permeability, pore-volume, or surface area of porous materials
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
    • G01N15/08—Investigating permeability, pore-volume, or surface area of porous materials
    • G01N2015/0866—Sorption

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Chemical & Material Sciences (AREA)
  • Dispersion Chemistry (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Investigating Or Analyzing Materials By The Use Of Fluid Adsorption Or Reactions (AREA)
  • Separation Of Gases By Adsorption (AREA)

Abstract

The invention discloses a method for predicting the methane gas adsorption performance of MOF based on a deep convolutional neural network, which predicts the methane gas adsorption performance of the expressed MOF by taking a CIF file storing an MOF basic three-dimensional structure as a data set of a data sample. The method comprises the steps of designing a classifier by using a convolutional neural network, converting an MOF basic three-dimensional structure in a CIF file into characteristics which can be accepted by the classifier, dividing the adsorption capacity of the MOF on methane gas into a plurality of intervals, and obtaining the prediction performance category of the MOF through model training.

Description

Method for predicting adsorption performance of MOF (metal organic framework) on methane gas based on deep convolutional neural network
Technical Field
The invention belongs to the technical field of intelligent prediction of functional materials, and particularly relates to a method for predicting the adsorption performance of MOF (metal organic framework) on methane gas based on a deep convolutional neural network.
Background
Metal-Organic Frameworks (MOFs), which are Organic-inorganic hybrid materials with intramolecular pores formed by self-assembly of Organic ligands and Metal ions or clusters through coordination bonds, belong to the category of coordination polymers. The metal-organic framework is composed of a three-dimensional periodic network, and building blocks, such as metal clusters and organic linkers, are composed of molecules. The possible combination of these numerous building blocks in different topologies results in an almost unlimited number of potential MOFs. In the MOFs, the arrangement of organic ligands and metal ions or clusters has a significant directionality, and different framework pore structures can be formed, thereby exhibiting different adsorption properties, optical properties, electromagnetic properties, and the like. MOFs present great development potential and attractive development prospects in modern materials science.
Maryam Pardakhti et al used the physical structure, the chemical structure and the combination of the physical and chemical structures of the metal organic framework as inputs, compared the prediction effects of the decision tree, poisson regression, support vector machine and random forest method, and reached the conclusion that the best prediction effect is obtained using the combination of the physical and chemical structures as inputs and using the random forest algorithm. However, in practical applications, the available metal organic frameworks are very numerous, and their chemical structures and some physical structures need to be obtained through calculation and a great deal of experiments, so that the method also has certain limitations in practical applications.
Disclosure of Invention
Aiming at the problem that the adsorption value of the MOF to the methane gas cannot be predicted by using the MOF three-dimensional basic structure information in the CIF file in the prior art, the invention provides a method for predicting the adsorption performance of the MOF to the methane gas based on a deep Convolutional Neural Network (CNN). The method is characterized in that the characteristic information in the basic space structure of the MOF is mined by using a deep learning model and combined, and the prediction of the methane adsorption value is realized by extracting the characteristics of the CIF file in which the basic space structure information of the MOF is stored.
The technical scheme is as follows:
a method for predicting the adsorption performance of MOF on methane gas based on a deep convolutional neural network comprises the following steps:
1) data preprocessing: expanding an original CIF file database and converting MOF basic three-dimensional structures in the CIF file into features acceptable for a classifier;
2) constructing a classifier by using a convolutional neural network, and carrying out iterative training on the classifier by dividing the methane gas adsorption capacity of the MOF into regions;
3) inputting the preprocessed data into a trained classifier and outputting the predicted adsorption value category of the MOF to the methane.
Further, the specific process of step 1) is as follows:
classifying an original data set according to the size of an adsorption value, randomly dividing the original data set into a training set and a testing set, and expanding data in other categories by taking one category with the largest data quantity as a reference; the MOF single unit cell in the CIF file rotates around a certain coordinate axis, the rotating angle is an included angle on a plane perpendicular to the coordinate axis, the side length of the MOF unit cell obtained after rotation is exchanged, the side length is reflected to the CIF file, namely the sides a, b and c in the file are exchanged, the atom coordinate is changed, the corresponding included angle of the a, b and c is changed, and therefore various CIF files generated based on the MOF with the same structure are obtained, and data expansion is achieved. And in addition, converting the atom relative coordinates stored in the data into absolute coordinates in space, counting the sizes of the atom absolute coordinates in all the MOF files, carrying out normalization processing on the atom absolute coordinates by using a maximum and minimum normalization method, and then amplifying the normalized MOF unit cell coordinates to obtain xy-plane, xz-plane and yz-plane projection matrixes.
Further, in the step 2), a classifier is constructed on the basis of a convolutional neural network and a residual block and is subjected to iterative training, and finally the classifier with the best classification effect is obtained through test of a test set, wherein a cross entropy function is used as a loss function of the classifier, and a LeakyRelu function is used as an activation function after each convolutional layer to prevent overfitting.
The invention has the following beneficial effects:
the method for predicting the adsorption performance of the MOF on the methane gas based on the deep convolutional neural network realizes data expansion on an MOF database by utilizing information such as angles in a CIF file, extracts the characteristics of a three-dimensional space basic structure of the MOF stored in the CIF file by utilizing the convolutional neural network, then performs classifier iterative training by utilizing the extracted characteristics, and finally applies the trained classifier to the predicted adsorption value of the MOF on the methane gas. The method can better process the CIF file storing the basic three-dimensional structure information of the MOF, extract the characteristics from the CIF file and obtain the gas prediction adsorption performance of the MOF on the methane with higher accuracy by using the characteristics.
Drawings
FIG. 1 is a flow chart of a method for predicting the adsorption performance of a MOF of the present invention to methane gas;
FIG. 2 is a ROC curve obtained by the classifier of the present invention;
FIG. 3 is a ROC curve obtained in the prior art.
Detailed Description
In order to make those skilled in the art better understand the technical solution of the present invention, the method for predicting the adsorption performance of MOF on methane gas based on deep convolutional neural network provided by the present invention is described in detail below with reference to the examples. The following examples are intended to illustrate the invention only and are not intended to limit the scope of the invention.
Step one, data preprocessing:
dividing the original data set into five types according to the size of the adsorption value, wherein the data set is divided into the following five types according to the weight ratio of 9: a scale of 1 is randomly divided from five classes into a training set and a test set. And then expanding the data in the other categories by taking the category with the largest data amount as a reference, wherein three included angles alpha, beta and gamma of the MOF single unit cell in three coordinate planes xy, xz and yz are stored in the CIF file, and the relative atomic coordinates of the atoms in the MOF single unit cell in the unit cell are stored in the CIF file. Since the cell after rotation also needs to be able to periodically expand in the x, y, and z directions, and the resulting MOF remains unchanged, it cannot be rotated at any angle. According to alpha, beta, gamma or other angles stored in the CIF file, anticlockwise rotation or clockwise rotation is carried out along the x axis, the y axis and the z axis, so that the MOF generated by the unit cell through periodic expansion in the x direction, the y direction and the z direction is obtained through rotation, the structure of the MOF is the same as that of the original MOF, namely the adsorption performance of the MOF on methane gas is kept unchanged, and information such as side length, angle, coordinates and the like in the CIF file is changed, so that a plurality of CIF files are generated on the MOF with the same structure, namely the expansion of an original data set is realized. Since the relative coordinates of the atoms in the MOF are stored in the CIF file, the relative coordinates of the atoms stored in the CIF file need to be converted into absolute coordinates in space after data expansion
Figure BDA0002479620450000031
In the above formula, x, y and z are absolute coordinates of atoms, x ', y ' and z ' are relative coordinates of atoms, a, b and c are three side lengths of a single unit cell of the MOF, and alpha, beta and gamma are included angles of the three side lengths of the MOF.
Then, the absolute coordinate sizes of atoms in all the MOF files are counted, normalization processing is carried out on the atomic coordinates, the normalization is carried out by using the maximum and minimum normalization, and then the normalized MOF unit cells are placed in a space with the size of 100x100x 100.
Figure BDA0002479620450000032
Figure BDA0002479620450000033
Figure BDA0002479620450000034
X in the above formulanew，ynew，znewIs an atomic coordinate in a 100x100x100 space, x, y, z are atomic absolute coordinates, xmax，ymax，zmaxIs the maximum coordinate value, x, of the atoms x, y, z in all datamin，ymin，zminThe minimum coordinate value of the atoms x, y and z in all data.
And finally, respectively projecting the MOF in the space of 100x100x100 x100 to an xy plane, a yz plane and an xz plane to obtain three projection matrixes.
Step two, designing a classifier: designing a classifier and performing iterative training on the basis of a convolutional neural network and a residual block, and finally testing by using a test set to obtain the classifier with the best classification effect, wherein a cross entropy function is used as a loss function of the classifier, and a LeakyRelu function is used as an activation function after each convolutional layer to prevent overfitting.
Step three, predicting the predicted adsorption category of the MOF to the methane gas by using a classifier: and inputting the preprocessed data into a classifier, and obtaining the predicted adsorption value category of the MOF to the methane according to the classification result output by the classifier.
Example 1
As shown in fig. 1, the method for predicting the adsorption performance of MOF using deep convolutional neural network on methane gas includes the following steps:
firstly, data preprocessing: the data set used In the examples was "In silicon discovery of metal-organic frames for precompoundation CO" published by Chung Y G et al 2016 In sciences Advances2The database obtained by capture using a genetic algorithm "contains 51163 MOFs. The data set contains 51163 MOF samples and the adsorption value of the MOF samples to methane gas, wherein each MOF sample contains the atom name, atom relative coordinates, the length of three sides of a unit cell and the angle of three sides. The adsorption value is counted up to 530cm3And/g, then classifying the MOFs according to the adsorption values, and dividing the MOFs into five types. After classification, according to 9: 1 into training and test sets, 0-106cm3The data of the/g interval are 7717 pieces, 106-212cm3The data of the/g interval are 13056 pieces in total, 212 and 318cm3The data of the/g interval are 16955 pieces, 318 and 424cm3The data of the/g interval are 7274, 424 and 530cm3The/g interval has 742 pieces of data. 318cm at 212-3And expanding training data of the rest intervals by taking the data amount of the/g interval as a reference. The data expansion method is to rotate the MOF clamp on a plane vertical to a certain coordinate axis around the coordinate axisThe angle is the angle, the relative coordinates of the atoms of the MOF after rotation are changed, but the spatial structure of the MOF is not changed, so that various CIF files are generated for the MOF with the same structure, namely, data expansion is realized. And after the data set is expanded, converting the atomic relative coordinates stored in the data into absolute coordinates in space, performing maximum and minimum normalization processing, and then projecting the coordinates after the normalization processing to an xy plane, an xz plane and a yz plane by enlarging by 100 times to obtain a 100x100 projection matrix.
II, designing a classifier: designing a classifier and performing iterative training on the basis of a convolutional neural network and a residual block, and finally testing by using a test set to obtain the classifier with the best classification effect, wherein a cross entropy function is used as a loss function of the classifier, and a LeakyRelu function is used as an activation function after each convolutional layer to prevent overfitting.
And thirdly, predicting the predicted adsorption category of the MOF on the methane gas by using a classifier: the invention inputs the preprocessed data into the classifier, and obtains the predicted adsorption value category of the MOF to the methane according to the classification result output by the classifier.
To demonstrate the effectiveness of the proposed method, we compared the ROC graph 2 of the data set on the classifier with the ROC graph 3 of the rest of the prior art.
ACC and AUC values of the CIF-CNN classifier model and the comparative experiment method created in the present invention are shown in Table 1:
TABLE 1
ACC AUC
CIF-CNN 82％ 0.94
SVM 29％ 0.54
DT 40％ 0.63
RF 46％ 0.76
As can be seen from table 1, on the data set, the proposed model prediction method using the deep convolutional network obtains the best ACC value and AUC value with a large difference, which indicates that the model prediction method using the deep convolutional network is more suitable for feature extraction of the MOF three-dimensional space infrastructure to obtain the predicted value of gas adsorption.
The present invention is not limited to the above-described examples, and various changes can be made without departing from the spirit and scope of the present invention within the knowledge of those skilled in the art.

Claims (3)

1. A method for predicting the adsorption performance of MOF on methane gas based on a deep convolutional neural network is characterized by comprising the following steps:
1) data preprocessing: expanding an original CIF file database and converting MOF basic three-dimensional structures in the CIF file into features acceptable for a classifier;
2) constructing a classifier by using a convolutional neural network, and carrying out iterative training on the classifier by dividing the methane gas adsorption capacity of the MOF into regions;
3) inputting the preprocessed data into a trained classifier and outputting the predicted adsorption value category of the MOF to the methane.
2. The method for predicting the adsorption performance of the MOF on the methane gas based on the deep convolutional neural network as claimed in claim 1, wherein the specific process of the step 1) is as follows:
classifying an original data set according to the size of an adsorption value, randomly dividing the original data set into a training set and a testing set, and expanding data in other categories by taking one category with the largest data quantity as a reference; the MOF single unit cell in the CIF file rotates around a coordinate axis, the rotating angle is an included angle on a plane vertical to the coordinate axis, the side length of the MOF unit cell obtained after rotation is interchanged, the side length is reflected to the side a, b and c in the CIF file and the atomic coordinate and the corresponding included angle of the a, b and c are correspondingly changed, so that various CIF file data generated based on the MOF with the same structure are expanded, in addition, the atomic relative coordinate stored in the data is converted into the absolute coordinate in the space, the atomic absolute coordinate size in all the MOF files is counted, the normalized MOF unit cell coordinate is amplified after the atomic absolute coordinate is normalized by using the maximum and minimum normalization method, and xy plane, xz plane and yz plane projection matrixes are obtained.
3. The method for predicting the adsorption performance of the MOF on the methane gas based on the deep convolutional neural network as claimed in claim 2, wherein in the step 2), a classifier is constructed on the basis of the convolutional neural network and a residual block, iterative training is carried out, and finally the classifier with the best classification effect is obtained by testing with a test set, wherein a cross entropy function is used as a loss function of the classifier, and a LeakyRelu function is used as an activation function after each convolutional layer to prevent overfitting.
CN202010374618.XA 2020-05-06 2020-05-06 Method for predicting methane gas adsorption performance of MOF (metal oxide film) based on deep convolutional neural network Active CN111755080B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010374618.XA CN111755080B (en) 2020-05-06 2020-05-06 Method for predicting methane gas adsorption performance of MOF (metal oxide film) based on deep convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010374618.XA CN111755080B (en) 2020-05-06 2020-05-06 Method for predicting methane gas adsorption performance of MOF (metal oxide film) based on deep convolutional neural network

Publications (2)

Publication Number Publication Date
CN111755080A true CN111755080A (en) 2020-10-09
CN111755080B CN111755080B (en) 2023-07-28

Family

ID=72673817

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010374618.XA Active CN111755080B (en) 2020-05-06 2020-05-06 Method for predicting methane gas adsorption performance of MOF (metal oxide film) based on deep convolutional neural network

Country Status (1)

Country Link
CN (1) CN111755080B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160334353A1 (en) * 2015-05-15 2016-11-17 General Electric Company Sensor for in situ selective detection of components in a fluid
US20190114544A1 (en) * 2017-10-16 2019-04-18 Illumina, Inc. Semi-Supervised Learning for Training an Ensemble of Deep Convolutional Neural Networks
CN110542710A (en) * 2019-09-16 2019-12-06 中国石油大学(华东) Preparation method of tungsten disulfide-based formaldehyde gas sensor and application of gas sensor in vehicle-mounted microenvironment detection

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160334353A1 (en) * 2015-05-15 2016-11-17 General Electric Company Sensor for in situ selective detection of components in a fluid
US20190114544A1 (en) * 2017-10-16 2019-04-18 Illumina, Inc. Semi-Supervised Learning for Training an Ensemble of Deep Convolutional Neural Networks
CN110542710A (en) * 2019-09-16 2019-12-06 中国石油大学(华东) Preparation method of tungsten disulfide-based formaldehyde gas sensor and application of gas sensor in vehicle-mounted microenvironment detection

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
GRACE ANDERSON ET AL.: "Attainable Volumetric Targets for Adsorption-Based Hydrogen Storage in Porous Crystals: Molecular Simulation and Machine Learning", 《THE JOURNAL OF PHYSICAL CHEMISTRY》, pages 120 - 129 *

Also Published As

Publication number Publication date
CN111755080B (en) 2023-07-28

Similar Documents

Publication Publication Date Title
Eldawy et al. CG_Hadoop: computational geometry in MapReduce
CN111914112B (en) Part CAD model reuse method based on point cloud classification network
CN104239299B (en) Three-dimensional model retrieval method and apparatus
Katz et al. Galaxy formation and the peaks formalism
Shimizu Tensor renormalization group approach to a lattice boson model
CN101699514B (en) Immune clone quantum clustering-based SAR image segmenting method
CN104008420A (en) Distributed outlier detection method and system based on automatic coding machine
CN106570104B (en) Multi-partition clustering preprocessing method for stream data
CN111950620A (en) User screening method based on DBSCAN and K-means algorithm
CN113886587A (en) Data classification method based on deep learning and map building method
Zhao et al. An application of spatial decision tree for classification of air pollution index
Shekhar et al. Topological data analysis enhanced prediction of hydrogen storage in metal–organic frameworks (MOFs)
CN111755080A (en) Method for predicting adsorption performance of MOF (metal organic framework) on methane gas based on deep convolutional neural network
CN102708589A (en) Three-dimensional target multi-viewpoint view modeling method on basis of feature clustering
Xianting et al. A density-based clustering algorithm for high-dimensional data with feature selection
CN104392128B (en) A kind of implementation method of three alloy nano particles catalyst stabilization structure
Qian et al. Structured sparse model based feature selection and classification for hyperspectral imagery
CN115620838A (en) Method for obtaining machine learning model combination for screening metal organic framework
CN114880690A (en) Source data time sequence refinement method based on edge calculation
CN114169220A (en) Based on phase-changeable Fe50Mn30Cr10Co10Atomic scale simulation method for phase change process of high-entropy alloy
Xiong et al. Data structure for on-lattice cluster–cluster aggregation model performance optimization
CN113838538A (en) Prediction method and system for organic-inorganic hybrid perovskite post-treatment amine based on random forest rapid classification
Kim et al. Self-similarity in fractal and non-fractal networks
Cheraghalizadeh et al. Gaussian free field in the background of correlated random clusters, formed by metallic nanoparticles
CN113298111A (en) Feature selection method for processing high-dimensional data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant