The content of the invention
The present invention provides a kind of intelligent design system and method for new material synthesis technique, by existing data in literature
Machine mould is trained and assessed after being pre-processed so that the new material synthesis technologic parameter of machine mould output magnanimity and prediction
The material property of the new material, substantially reduces the construction cycle of new material, and to provide intelligence excellent for the synthesis of new material
The synthesis technique of change.
In order to solve above-mentioned technical problem, the technical solution used in the present invention is:
One aspect of the present invention provides a kind of intelligent design system of new material synthesis technique, including data in literature collection mould
Block, data in literature processing module, machine training and evaluation module, new material design module, new material performance test module;
Data in literature acquisition module is used for by data mining technology collection comprising raw material composition of vector and synthetic material
The data in literature of performance vectors, constitutes data in literature collection;
Data in literature processing module is used for the data in literature collection to gathering and pre-processes, and obtains the corresponding work of synthetic material
Skill parameter and its performance parameter, form machine learning data set;
Machine is trained to be used to for machine learning data set to be divided into machine training dataset with machine assessment with evaluation module
Data set;Machine training dataset is input into into machine mould, machine mould is trained, and data set is assessed with machine
The performance of the machine mould that assessment has been trained;
New material design module is used for using the machine mould design new material synthesis technologic parameter trained and assess, and
The material property parameter of prediction new material;
New material performance test module is used for the material property parameter sieve of the new material that module prediction is designed according to new material
Selecting needs the new material of synthesis, by the new material filtered out described in experiment synthesis, then the new material that experiment is obtained is carried out
Material properties test, calculates the material property parameter predicted value of the new material of new material design module prediction and the institute for testing synthesis
State the error between the material property parameter measured value of new material;
Machine is trained and evaluation module is additionally operable to new material designed using machine mould and through experiment test and is synthesized
The material property of technological parameter and new material carries out retraining to machine mould.
Further, the data in literature collection for gathering is pre-processed by the method for artificial mark, obtains the synthesis
The corresponding technological parameter of material and its performance parameter, constitute the machine learning data trained for machine mould and assess.
Further, the machine mould be based on the machine mould of artificial neural network, by input layer, hidden layer and
Output layer is constituted.
Specifically, the training of the machine mould includes forward-propagating process and back-propagation process, the forward-propagating
The synthetic material technological parameter that the input layer input machine training data of process is concentrated, the synthetic material technological parameter Jing implies
After layer is processed, output layer, output layer output synthetic material performance parameter predicted value are transmitted to;The back-propagation process is according to machine
The actual input of model and output, using gradient descent method and each layer weighting parameter of interative computation computing machine model.
Specifically, when the machine assessment data set assesses the performance of training machine model, input layer input machine is commented
Estimate the synthetic material technological parameter in data set, the synthetic material technological parameter is transmitted to output layer Jing after hidden layer process, defeated
Go out layer output synthetic material performance parameter predicted value.
Further, the materials synthesis technological parameter includes:Raw material ratio parameter, temperature parameter, pressure parameter,
Chemical time parameter;The material property parameter includes physical function parameter, chemical property parameter.
Another aspect of the present invention provides a kind of Intelligentized design method of new material synthesis technique, it is characterised in that:Including such as
Lower step:
Data in literature is gathered according to raw material composition of vector and synthetic material performance vectors;
Obtain the corresponding technological parameter of synthetic material and its performance parameter;
Machine mould is trained, and assesses the performance of the machine mould trained;
The machine mould design new material synthesis technologic parameter trained and assessed, and predict the material property ginseng of new material
Number;
Filtering out needs the new material of synthesis, and by the experiment synthesis new material, then material is carried out to the new material
Material performance test;
Retraining is carried out to machine mould.
A kind of intelligent design system and method for new material synthesis technique that the present invention is provided, compared with prior art, this
Data in literature pretreatment of the invention by the method for artificial mark to gathering, obtains the corresponding technological parameter of synthetic material and its property
Energy parameter is simultaneously trained and assessment machine mould, is joined using the machine mould output magnanimity new material synthesis technique trained and assess
The material property of the new material is counted and predicts, filtering out according to the material property of the new material of machine mould prediction needs synthesis
New material, and by the material property of new material described in experimental evaluation whether meet machine mould prediction material property, from
And the new material for meeting material performance requirement is obtained, the big number of the relevant material technology synthesis that the present invention is recorded based on existing document
Factually show the foundation and application of machine mould so that the design of the new material synthesis technique is not only more intelligent, also causes to be obtained
The new material synthesis technique for taking more optimizes, and substantially reduces the construction cycle of new material, solves in prior art by repeatedly
Adjust the parameter of new material synthesis technique and carry out numerous experiments and be unsuccessfully possible to obtain the new material for meeting performance requirement
Problem.
Specific embodiment
Embodiments of the present invention are specifically illustrated below in conjunction with the accompanying drawings, and accompanying drawing is only for reference and explanation is used, and it is right not constitute
The restriction of scope of patent protection of the present invention.
On the one hand the embodiment of the present invention provides a kind of intelligent design system of new material synthesis technique, as shown in Fig. 2 including
Data in literature acquisition module, data in literature processing module, machine training and evaluation module, new material design module, new material
Can test module;
Data in literature acquisition module is used for by data mining technology collection comprising raw material composition of vector A and synthetic material
The data in literature of performance vectors B, constitutes data in literature collection C;Wherein, the raw material composition of vector A is raw material composition composition
Vector;The data in literature of the collection is the document number for including raw material composition composition of vector and synthetic material performance vectors simultaneously
According to;In the prior art, the exploitation of new material is based on the experiment based on trial and error, therefore, having had in a large number can be for reference
Materials synthesis technique data in literature, the data in literature constitutes new material synthesis technique intelligent design system of the present invention
Big data;
In embodiments of the present invention, data in literature collection of the data in literature processing module by the method for artificial mark to collection
C is pre-processed, and obtains the corresponding technological parameter of synthetic material and its performance parameter, constitutes machine learning data set D;
Machine is trained to be used to for the machine learning data set to be divided into machine training dataset E and machine with evaluation module
Assessment data set F;Machine training dataset E is input into into machine mould, machine mould is trained, and assessed with machine
The performance of the machine mould that data set F assessments have been trained;The machine learning data set generally presses 7:3 relative size is divided at random
With being that machine training dataset and machine assess data set;
As shown in figure 3, being machine mould structural representation, the machine mould is based on the machine mould of artificial neural network
Type, is made up of input layer f1, hidden layer f2 and output layer f3, and per layer includes several neurons.
The training of the machine mould includes forward-propagating process and back-propagation process, and the machine mould is previously provided with
Error preset value a, during the forward-propagating of machine mould, the synthetic material work that input layer input machine training data is concentrated
Skill parameter xi, synthetic material technological parameter xi is transmitted to output layer, output layer output synthetic material performance ginseng Jing after hidden layer process
Number predicted value y, calculates error b between predicted value y and measured value g of the machine mould acquisition, wherein, the measured value is
The corresponding performance parameter of synthetic material technological parameter that machine training data is concentrated;If error b is more than error preset value a,
The backpropagation of machine mould is then proceeded to, error signal is returned along original connecting path, change the weights of each layer neuron,
So that error b is less than error preset value a.The input of each neuron of the machine mould is as follows with output relation formula:
Wherein, wi is the weight coefficient of neuron;θ is the threshold value of neuron, and f (X) is the excitation function of neuron;It is each
The state of layer neuron only affects the state of next layer of neuron.
In the back-propagation process of the machine mould, according to the actual input of machine mould and output, using under gradient
Drop method and each layer weighting parameter of interative computation computing machine model, complete the training of machine mould.
When the machine assessment data set assesses the performance of training machine model, input layer input machine assessment data set
In synthetic material technological parameter xi, the synthetic material technological parameter xi Jing hidden layer process after, be transmitted to output layer, output layer
Output synthetic material performance parameter predicted value y, the error between the predicted value and measured value of the machine mould output trained
During more than error preset value a, then adjust machine mould in hidden layer the number of plies and neuron number, using machine training data
Set pair machine mould re -training, and the performance that data set assesses again the machine mould trained is assessed using machine, until
Error between the predicted value and measured value of the machine mould output trained is less than or equal to error preset value a, wherein, the reality
Measured value is the corresponding synthetic material performance parameter of synthetic material technological parameter that machine is assessed in data set.
New material design module is used to design the new material synthesis technique of magnanimity using the machine mould trained and assess
Parameter, and predict the material property parameter of the new material;
New material performance test module be used for according to new material design module prediction new material material property parameter from
The new material for needing synthesis is filtered out in the magnanimity new material synthesis technologic parameter of the machine mould design trained and assessed, is led to
The new material filtered out described in experiment synthesis is crossed, then material properties test is carried out to the new material that experiment is obtained;Calculate new material
The material property parameter predicted value of the new material of design module prediction and the material property parameter of the new material of experiment synthesis
Error between measured value;
Machine is trained and evaluation module is additionally operable to new material designed using machine mould and through experimental evaluation and is synthesized
The material property of technological parameter and correspondence new material carries out retraining to machine mould.
Wherein, if the material property parameter predicted value of the new material of new material design module prediction is described with experiment synthesis
Error between the material property parameter measured value of new material is less than error preset value a, then the new material for obtaining the experiment
Material property parameter and its corresponding synthesis technologic parameter machine mould is carried out after retraining, terminate experiment;If new material
The material property parameter predicted value of the new material of design module prediction and the material property parameter of the new material of experiment synthesis
Error between measured value is more than error preset value a, the material property parameter and its correspondence of the new material that the experiment is obtained
Synthesis technologic parameter machine mould is carried out after retraining, the synthesis technologic parameter for redesigning new material simultaneously predicts new material
Material property, until experiment is obtained when meeting the new material of material performance requirement, terminate experiment;
Preferably, the data in literature collection for gathering is pre-processed by the method for artificial mark, obtains the synthesis material
Expect corresponding technological parameter and its performance parameter, constitute the machine learning data trained for machine mould and assess.
Preferably, the machine mould is based on the machine mould of artificial neural network, by input layer, hidden layer and output
Layer composition.
Preferably, the training of the machine mould includes forward-propagating process and back-propagation process, the forward-propagating
The synthetic material technological parameter that the input layer input machine training data of process is concentrated, the synthetic material technological parameter Jing implies
After layer is processed, output layer, output layer output synthetic material performance parameter predicted value are transmitted to;The back-propagation process is according to machine
The actual input of model and output, using gradient descent method and each layer weighting parameter of interative computation computing machine model.
Preferably, when the machine assessment data set assesses the performance of training machine model, input layer input machine is commented
Estimate the synthetic material technological parameter in data set, the synthetic material technological parameter is transmitted to output layer Jing after hidden layer process, defeated
Go out layer output synthetic material performance parameter predicted value.
Specifically, the materials synthesis technological parameter is included but is not limited to:Raw material ratio parameter, temperature parameter, pressure
Parameter, chemical time parameter;The material property parameter includes but is not limited to physical function parameter, chemical property parameter.
On the other hand the embodiment of the present invention provides a kind of Intelligentized design method of new material synthesis technique, including following step
Suddenly:
Data in literature C is gathered according to raw material composition of vector A and synthetic material performance vectors B;
From data in literature C of collection, the corresponding technological parameter of synthetic material and its performance parameter are obtained;
Machine mould is trained, and assesses the performance of the machine mould trained;
New material synthesis technologic parameter is designed with the machine mould trained and assess, and predicts the material property of new material
Parameter;
Filtering out needs the new material of synthesis, and by the experiment synthesis new material, then material is carried out to the new material
Material performance test;
Retraining is carried out to machine mould.
A kind of Intelligentized design method flow chart of new material synthesis technique is illustrated in figure 4, in the present embodiment, is led to
Data in literature of the data mining technology collection comprising raw material composition of vector A with synthetic material performance vectors B is crossed, document number is constituted
According to collection C;Wherein, the raw material composition of vector A is raw material composition composition of vector;The data in literature of the collection is bag simultaneously
The data in literature of the composition of vector of composition containing raw material and synthetic material performance vectors;
The data in literature collection C for gathering is pre-processed by the method for artificial mark, obtains the corresponding work of synthetic material
Skill parameter and its performance parameter, constitute machine learning data set D;The machine learning data set is divided into into machine training data
Collection E assesses data set F with machine;And machine training dataset E is input into into machine mould, machine mould is trained,
The performance of the machine mould that data set F assessments have been trained is assessed with machine;The machine learning data set generally presses 7:3 it is relative
Size is randomly assigned to be that machine training dataset and machine assess data set;
As shown in figure 3, being machine mould structural representation, the machine mould is based on the machine mould of artificial neural network
Type, is made up of input layer f1, hidden layer f2 and output layer f3, and per layer includes several neurons.
The training of the machine mould includes forward-propagating process and back-propagation process, and the machine mould is previously provided with
Error preset value a, during the forward-propagating of machine mould, the synthetic material work that input layer input machine training data is concentrated
Skill parameter xi, synthetic material technological parameter xi is transmitted to output layer, output layer output synthetic material performance ginseng Jing after hidden layer process
Number predicted value y, calculates error b between predicted value y and measured value g of the machine mould acquisition, wherein, the measured value is
The corresponding performance parameter of synthetic material technological parameter that machine training data is concentrated;If error b is more than error preset value a,
The backpropagation of machine mould is then proceeded to, error signal is returned along original connecting path, change the weights of each layer neuron,
So that error b is less than error preset value a.The input of each neuron of the machine mould is as follows with output relation formula:
Wherein, wi is the weight coefficient of neuron;θ is the threshold value of neuron, and f (X) is the excitation function of neuron;It is each
The state of layer neuron only affects the state of next layer of neuron.
In the back-propagation process of the machine mould, according to the actual input of machine mould and output, using under gradient
Drop method and each layer weighting parameter of interative computation computing machine model, complete the training of machine mould.
When the machine assessment data set assesses the performance of training machine model, input layer input machine assessment data set
In synthetic material technological parameter xi, the synthetic material technological parameter xi Jing hidden layer process after, be transmitted to output layer, output layer
Output synthetic material performance parameter predicted value y, the error between the predicted value and measured value of the machine mould output trained
During more than error preset value a, then adjust machine mould in hidden layer the number of plies and neuron number, using machine training data
Set pair machine mould re -training, and the performance that data set assesses again the machine mould trained is assessed using machine, until
Error between the predicted value and measured value of the machine mould output trained is less than or equal to error preset value a, wherein, the reality
Measured value is the corresponding synthetic material performance parameter of synthetic material technological parameter that machine is assessed in data set.
A kind of Intelligentized design method of new material synthesis technique also includes designing the new of module prediction according to new material
The material property parameter of material is screened from the magnanimity new material synthesis technologic parameter of the machine mould design trained and assess
Go out to need the new material of synthesis, by the new material filtered out described in experiment synthesis, then material is carried out to the new material that experiment is obtained
Material performance test;The material property parameter predicted value for calculating the new material of new material design module prediction is described with experiment synthesis
Error between the material property parameter measured value of new material;
Wherein, if the material property parameter predicted value of the new material of new material design module prediction is described with experiment synthesis
Error between the material property parameter measured value of new material is less than error preset value a, then the new material for obtaining the experiment
Material property parameter and its corresponding synthesis technologic parameter machine mould is carried out after retraining, terminate experiment;If new material
The material property parameter predicted value of the new material of design module prediction and the material property parameter of the new material of experiment synthesis
Error between measured value is more than error preset value a, the material property parameter and its correspondence of the new material that the experiment is obtained
Synthesis technologic parameter machine mould is carried out after retraining, the synthesis technologic parameter for redesigning new material simultaneously predicts new material
Material property, until experiment is obtained when meeting the new material of material performance requirement, terminate experiment;And designed using machine mould
And through experimental evaluation new material synthesis technologic parameter and correspondence new material material property machine mould is instructed again
Practice.
Above disclosed is only presently preferred embodiments of the present invention, it is impossible to the rights protection model of the present invention is limited with this
Enclose, therefore the equivalent variations made according to scope of the present invention patent, still belong to the scope that the present invention is covered.