CN109346135A - A method of hydrone energy is calculated by deep learning - Google Patents
A method of hydrone energy is calculated by deep learning Download PDFInfo
- Publication number
- CN109346135A CN109346135A CN201811133773.1A CN201811133773A CN109346135A CN 109346135 A CN109346135 A CN 109346135A CN 201811133773 A CN201811133773 A CN 201811133773A CN 109346135 A CN109346135 A CN 109346135A
- Authority
- CN
- China
- Prior art keywords
- energy
- matrix
- layer
- group
- configuration
- 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.)
- Pending
Links
Landscapes
- Organic Low-Molecular-Weight Compounds And Preparation Thereof (AREA)
Abstract
The invention discloses a kind of methods for calculating hydrone energy by deep learning, belong to molecular energy computing technique field, including S1: building moisture subdata base, the moisture subdata base include 1000 various configuration hydrones space coordinate and energy corresponding with configuration;S2: m configuration and corresponding energy are randomly selected as training group, remaining 1000-m configuration is as test group;S3: being two hydrogen-oxygen key bond distance r by training group and fractions tested subspace coordinate transformationO‑H1, rO‑H2, molecule bond angle θ and three interatomic distances 1/r reciprocalO‑H1,1/rO‑H2,1/rH1‑H2;S4: training group energy datum is extracted as output energy matrix;S5: building test group structure parameters input matrix and output energy matrix;S6: structure is calculated using double nervous layers, energy is learnt;The ratio of training group and test group of influence present invention decreases to(for) training result accuracy.
Description
Technical field
The present invention relates to molecular energy computing technique fields, and in particular to a kind of to calculate hydrone energy by deep learning
Method.
Background technique
The machine learning of contemporary artificial intelligence, in molecular structure optimization, in minimum energy calculating field, distinguishing feature be with
The increase of initial matrix freedom degree, i.e. molecular structure more complex free degree it is bigger, it is necessary to increase the ratio of training group and test group
Value, can just obtain accurate result.It is restricted by this feature, this method is in complicated molecule system or polymolecular system application
In, hardly result in satisfactory result.
Summary of the invention
To solve defect existing in the prior art, the present invention, which provides, a kind of calculates hydrone energy by deep learning
Method reduces influence of the ratio of training group and test group for training result accuracy.
The present invention is that technical solution used by solving its technical problem is: one kind calculating water molecule energy by deep learning
The method of amount, which comprises the steps of:
S1: building moisture subdata base, the moisture subdata base include that the space of 1000 various configuration hydrones is sat
Mark and energy corresponding with configuration;
S2: m configuration and corresponding energy are randomly selected and is made as training group, remaining 1000-m configuration and corresponding energy
For test group;
S3: being two hydrogen-oxygen key bond distances, molecule bond angle θ, three interatomic distances by training group molecule space coordinate transformation
1/r reciprocalO-H1,1/rO-H2,1/rH1-H2, described two hydrogen-oxygen key bond distances are respectively rO-H1, rO-H2;And with this six parameters
Form the structure parameters input matrix of training group;
S4: training group energy datum is extracted as output energy matrix, and correspond with structure parameters input matrix, i.e.,
The structure parameters input matrix R and output energy matrix E of training grouprealIt is respectively as follows:
S5: building test group structure parameters input matrix and output energy matrix, matrix line number are 1000-m;Test
The structure parameters input matrix R* and output energy matrix E of groupreal*It is respectively as follows:
S6: according to training group structure parameters, structure is calculated using double nervous layers, hydrone energy is learnt, by double
Water molecule energy moment matrix E is calculated in nervous layercalc:
Ecalc=[tf.nn.relu (R × Win+bin)]×Wout+bout;
Wherein tf.nn.relu is line rectification function, and R is the structure parameters input matrix of training group, WinFor first nerves
Layer weight matrix, binFor first nerves layer bias matrix, WoutFor nervus opticus layer weight matrix, boutFor the biasing of nervus opticus layer
Matrix.
Further, the specific steps of the step S6 are as follows: first nerves layer uses relu activation primitive, nervus opticus layer
Any activation primitive is not used, and every layer of neuron number is 10, first nerves layer weight matrix WinAre as follows:
First nerves layer bias matrix binAre as follows:
Nervus opticus layer weight matrix WoutAre as follows:
Nervus opticus layer bias matrix boutAre as follows:
Further, WinAnd WoutInitial value is generated by random number, binAnd boutInitial value is all set to 0.1.
The beneficial effects of the present invention are: solve the problems, such as that traditional calculations chemical method relies on CPU monokaryon dominant frequency, it will be current
Newest GPU computing technique is applied to the molecular energy calculating field of traditional calculations chemistry, to get rid of traditional calculations platform
It restricts;Convergence problem is avoided, computational efficiency is optimal level, reduces the ratio of training group and test group for training
As a result the influence of accuracy.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with embodiment, to skill of the invention
Art scheme carries out clear, complete description.
Embodiment 1
A method of hydrone energy is calculated by deep learning, is sat using the space of 1000 various configuration hydrones
Mark and corresponding energy.Two hydrogen-oxygen key variation ranges of hydrone areBond angle variation range
It is 104.2 ± 8.59 °.Directly with molecular configuration training molecular energy, m configuration and corresponding energy are randomly selected as training
Group, remaining 1000-m are used as test group, verify the correctness of training result.
It is two hydrogen-oxygen key bond distance (r by training group molecule space coordinate transformationO-H1, rO-H2), molecule bond angle (θ) and three
Inverse (the 1/r of a interatomic distanceO-H1,1/rO-H2,1/rH1-H2), and the defeated of training group structure parameters is formed with this six parameters
Enter matrix.Training group energy datum is extracted as output energy matrix, and correspond with structure parameters input matrix, that is, trained
The structure parameters input matrix R and output energy matrix E of grouprealIt is respectively as follows:
Test group structure parameters input matrix and output energy matrix are constructed, matrix line number is 1000- m;That is test group
Structure parameters input matrix R* and output energy matrix Ereal*It is respectively as follows:
According to training group structure parameters, structure is calculated using double nervous layers, energy is learnt.To guarantee learning efficiency,
First nerves layer uses relu activation primitive, and nervus opticus layer does not use any activation primitive, and every layer of neuron number is 10,
Energy matrix E is calculated by double nervous layerscalc。
Ecalc=[tf.nn.relu (R × Win+bin)]×Wout+bout (3)
Here, tf.nn.relu is line rectification function;R is structure parameters matrix;WinFor first nerves layer weight matrix,
binFor first nerves layer bias matrix,
Nervus opticus layer weight matrix WoutAre as follows:
Nervus opticus layer bias matrix boutAre as follows:Wherein, WinAnd WoutInitial value is generated by random number, bin
And boutInitial value is all set to 0.1;EcalcFor calculated energy value:
Embodiment 2
This gives the preferred embodiments of hardware platform of the invention and software environment.
Select i5-6500CPU@3.20GHz/NVIDIA Corporation GK208 [the GeForce GT of low side
730]/4G Mem hardware platform, to obtain higher universal performance;Software environment is Linux kernel 4.9/
TensorFlow-GPU 1.8.0 (is installed) by pip mode, and driver is CUDA 9.0/cuDNN 7.1.
Embodiment 3
This gives the preferred embodiments that input data of the present invention is chosen.
Using the water data set of the offers such as Brockherde, the data set include 1000 hydrones configuration and with
One-to-one energy, configuration using Bohrpositions express, energy unit kcal/mol.This method is ensuring
ML-OF method calculates molecular energy accuracy as training set increases in the case where raising, avoids using gradient descent method meter
It calculates and minimizes gross energy, result (PBE) institute calculated result approximate with standard DFT is used is compared.Hydrone parameter setting
For three: two bond distances and a bond angle.According to PBE result building optimization hydrone configuration (θ0=
104.2 °) it is trained starting point,And uniform configuration is generated between ± 8.59 °.
Embodiment 4
This gives the preferred embodiments that result of the present invention calculates.
Using tensorflow-gpu default configuration, inactive CPU concurrent operation.Four thread i5CPU of double-core, every thread account for
It is about 40% or so with rate, every thread memory usage is about 19.4%.Default opens GPU operation, and memory clock frequency is
0.9015GHz, memory use 1.93/1.95GiB.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Anyone skilled in the art within the technical scope of the present disclosure, according to the technique and scheme of the present invention and its
Inventive concept is subject to equivalent substitution or change, should be covered by the protection scope of the present invention.
Claims (3)
1. a kind of method for calculating hydrone energy by deep learning, which comprises the steps of:
S1: building moisture subdata base, the moisture subdata base include 1000 various configuration hydrones space coordinate and
Energy corresponding with configuration;
S2: m configuration and corresponding energy are randomly selected as training group, remaining 1000-m configuration and corresponding energy are as test
Group;
S3: being that two hydrogen-oxygen key bond distances, molecule bond angle θ, three interatomic distances fall by training group molecule space coordinate transformation
Number 1/rO-H1,1/rO-H2,1/rH1-H2, described two hydrogen-oxygen key bond distances are respectively rO-H1, rO-H2;And it is formed with this six parameters
The structure parameters input matrix of training group;
S4: training group energy datum is extracted as output energy matrix, and correspond with structure parameters input matrix, that is, trained
The structure parameters input matrix R and output energy matrix E of grouprealIt is respectively as follows:
S5: building test group structure parameters input matrix and output energy matrix, matrix line number are 1000-m;That is the structure of test group
Shape parameter input matrix R* and output energy matrix Ereal*It is respectively as follows:
S6: according to training group structure parameters, structure is calculated using double nervous layers, hydrone energy is learnt, by double nerves
Water molecule energy moment matrix E is calculated in layercalc:
Ecalc=[tf.nn.relu (R × Win+bin)]×Wout+bout;
Wherein tf.nn.relu is line rectification function, and R is the structure parameters input matrix of training group, WinFor first nerves layer power
Weight matrix, binFor first nerves layer bias matrix, WoutFor nervus opticus layer weight matrix, boutSquare is biased for nervus opticus layer
Battle array.
2. a kind of method for calculating hydrone energy by deep learning according to claim 1, which is characterized in that described
The specific steps of step S6 are as follows: first nerves layer uses relu activation primitive, and nervus opticus layer does not use any activation primitive, often
Layer neuron number is 10,
First nerves layer weight matrix WinAre as follows:
First nerves layer bias matrix binAre as follows:
Nervus opticus layer weight matrix WoutAre as follows:
Nervus opticus layer bias matrix boutAre as follows:
3. a kind of method for calculating hydrone energy by deep learning according to claim 2, which is characterized in that WinWith
WoutInitial value is generated by random number, binAnd boutInitial value is all set to 0.1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811133773.1A CN109346135A (en) | 2018-09-27 | 2018-09-27 | A method of hydrone energy is calculated by deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811133773.1A CN109346135A (en) | 2018-09-27 | 2018-09-27 | A method of hydrone energy is calculated by deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109346135A true CN109346135A (en) | 2019-02-15 |
Family
ID=65306803
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811133773.1A Pending CN109346135A (en) | 2018-09-27 | 2018-09-27 | A method of hydrone energy is calculated by deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109346135A (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6587845B1 (en) * | 2000-02-15 | 2003-07-01 | Benjamin B. Braunheim | Method and apparatus for identification and optimization of bioactive compounds using a neural network |
CN1886659A (en) * | 2003-10-14 | 2006-12-27 | 维颂公司 | Method and apparatus for analysis of molecular configurations and combinations |
US20170329892A1 (en) * | 2016-05-10 | 2017-11-16 | Accutar Biotechnology Inc. | Computational method for classifying and predicting protein side chain conformations |
-
2018
- 2018-09-27 CN CN201811133773.1A patent/CN109346135A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6587845B1 (en) * | 2000-02-15 | 2003-07-01 | Benjamin B. Braunheim | Method and apparatus for identification and optimization of bioactive compounds using a neural network |
CN1886659A (en) * | 2003-10-14 | 2006-12-27 | 维颂公司 | Method and apparatus for analysis of molecular configurations and combinations |
US20170329892A1 (en) * | 2016-05-10 | 2017-11-16 | Accutar Biotechnology Inc. | Computational method for classifying and predicting protein side chain conformations |
Non-Patent Citations (7)
Title |
---|
FELIX BROCKHERDE等: "By-passing the Kohn-Sham equations with machine learning", 《NATURE COMMUNICATIONS》 * |
GREGOIRE MONTAVON等: "Learning Invariant Representations of Molecules for Atomization Energy Prediction", 《ANNUAL CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS》 * |
KRISTOF T.SCHUTT等: "Quantum-chemical insights from deep tensor neural networks", 《NATURE COMMUNICATIONS》 * |
刘芹: "应用神经网络方法优化密度泛函近似中的半经验参数", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
陈俊等: "基于神经网络的分子体系势能面构造", 《中国科学:化学》 * |
陈柳杨: "构建多维化学反应势能面新方法", 《中国优秀硕士学位论文全文数据库工程科技I辑》 * |
陈炽宏: "基于神经网络的大分子体系势能面的构建", 《中国优秀硕士学位论文全文数据库工程科技I辑》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Hannam et al. | Simple model of complete precessing black-hole-binary gravitational waveforms | |
CN106842113B (en) | Direction of arrival high-precision estimation method in the case of height 1 bit quantization of sampling | |
CN106099393A (en) | Submatrix bare cloth based on genetic algorithm and array element sparse antenna array are comprehensively structured the formation method | |
WO2019033820A1 (en) | Optimization model for quick boundary searching for static voltage stability region of electric power system | |
CN105445696A (en) | Nested L-shaped antenna array structure and direction of arrival estimation method thereof | |
CN109946701A (en) | A kind of cloud coordinate transformation method and device | |
CN102297687A (en) | Calibrating method for electronic compass | |
CN110083920A (en) | The analysis method of non-proportional damping structural random response under a kind of geological process | |
CN106154217B (en) | The method for self-calibrating eliminated when mutual coupling is unknown in ULA and UCA based on spatial spectrum puppet peak | |
CN103020471B (en) | Block Ritz vector generation method for fluctuating wind-induced response calculation of long-span roof structure | |
CN106602282A (en) | Mutual-coupling-contained time modulation array synthesis method based on difference evolution algorithm | |
Duan et al. | A study of the mixed layer of the South China Sea based on the multiple linear regression | |
CN109346135A (en) | A method of hydrone energy is calculated by deep learning | |
Lou et al. | Multiobjective equivalent static wind loads on complex tall buildings using non-gaussian peak factors | |
CN106546957A (en) | A kind of ultra-short baseline localization method for the polynary battle array of arbitrary plane | |
CN109411028A (en) | The method for calculating hydrone energy based on molecular freedom deep learning | |
Amitai et al. | The effect of wind-stress over the Eastern Mediterranean on deep-water formation in the Adriatic Sea | |
Daniela et al. | Land-based remote sensing of coastal basins: Use of an HF radar to investigate surface dynamics and transport processes in the Gulf of Naples | |
Duan et al. | Efficient computational algorithms for approximate optimal designs | |
CN103065363A (en) | Method of fast importing pole tower three-dimensional model to scene | |
CN109147877A (en) | A method of ethane molecule energy is calculated by deep learning | |
CN103438905B (en) | A kind of star sensor star catalogue complete evaluation method | |
CN108521128B (en) | Rapid search method for static voltage security domain boundary of electric power system | |
CN106060921B (en) | A kind of indoor location based on manifold learning determines method and device | |
Malik et al. | Optimal matrix partitioning for data parallel computing on hybrid heterogeneous platforms |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190215 |
|
RJ01 | Rejection of invention patent application after publication |