A neural network-assisted method for the synthesis of chemical materials

Abstract

The invention discloses a method for assisting the synthesis of chemical materials by a neural network, comprising the following steps: step 1, collecting chemical reaction data, removing useless attributes and non-numeric fields, and performing one-hot encoding; step 2, removing relatively strong correlation Redundant data; Step 3, perform feature engineering and standardization on the data; Step 4, use a variety of methods to reduce the dimension of the data set, screen out the method with better dimensionality reduction effect, and use the dimensionality reduction data as the data set ; Step 5, extract the local features and all features of the data set, and use them as the input of the neural network after fusion; Step 6, build a neural network architecture, and train to obtain a chemical material crystallization prediction model; Step 7, predict the chemical reaction to be predicted; the present invention Using the algorithm model to reversely select the feature data and establish a prediction model can more accurately predict the crystallization of the chemical reaction, and further assist the synthesis of the chemical reaction.

Description

technical field [0001] The invention belongs to the technical field of new material research and development in chemical material engineering, and relates to a method for synthesizing chemical materials assisted by a neural network. Background technique [0002] Material innovation is the foundation and driving force for technological progress and industrial development. Traditional research and development of new materials often uses the trial-and-error method to obtain expected materials. This method has cumbersome experimental steps, long research and development cycles, and waste of resources, and often fails to meet experimental expectations during operation. , resulting in a large amount of unsatisfactory data, which complicates data processing; with the development of computer technology, many theories for calculating the structure and properties of materials have emerged one after another, such as first-principles calculation, field simulation, finite element analysis...

AI Technical Summary

Problems solved by technology

[0006] At present, through feature selection technology, it has been proved that a two-parameter model can predict whether a given molecule can be crystallized, and its accuracy is as high as 80%. However, most existing technologies use a single machine learning algorithm for data dimensionality reduction, such as PCA, LDA, etc., but a single algorithm has certain defects: LDA is not suitable for dimensionality reduction of non-Gaussian distribution sample data, and the meaning of each feature dimension has a certain degree of ambiguity when PCA dimensionality reduction is used, which is not as strong as the interpretation of the original sample features. , the non-principal components with small variance may also contain important information of sample differences, and the loss of dimensionality reduction may have an impact on subsequent data processing, so the accuracy of using it for chemical material crystallization prediction is not high

Images