Machine Learning Study On The Electron Emission Properties Of The Thermionic Impregnants

Nowadays,human beings are in the highly developed information technology era,in which all industries are in a rapid development stage.Vacuum electronics can play a very critical role in military defense,satellite communications,civil navigation and many other fields,and thus has become one of the hot spots for domestic and international research.As an electron source commonly used in vacuum electronics,thermionic cathodes are naturally the focus of scientific research.In order to meet the trend of vacuum electronics towards miniaturization,high frequency and high power,the development of thermionic cathodes with large emission current density is the primary research goal in the field of thermionic cathodes.Traditional materials research methods are often cumbersome and inefficient,making it difficult to meet the needs of today’s society.In recent years,with the generation of large amounts of data in various fields,machine learning,a new data processing technique,has gradually been applied in various fields,and in the field of materials science,there have been reports of using machine learning to develop new materials or assist in conducting research.Thesis will analyze the emission data of thermionic cathodes and predict the emission current density of thermionic cathodes based on the regression method in machine learning in order to shorten the development process of thermionic cathode materials and improve the development efficiency by expecting to predict the combination of input features with large emission.The main research of thesis is summarized as follows:1.Collect thermionic cathode emission data from published papers,analyze the data and separate features and labels,use the emission current density as the final label to be predicted,the chemical composition and the rest of the features as input features,and organize the data to form a data set.2.Invoke Linear Regression,Random Forest Regressor,Extreme Gradient Boosting(XGBoost)and Gradient Boosting Regression Tree and other machine learning regression algorithms and integrated algorithms to build models,substitute the training set in the established dataset into the model for training,and predict the emission current density of the test set,compare the prediction results of each algorithm,and get the gradient boosting regression tree algorithm has the best prediction performance.3.In order to solve the problem of small amount of thermionic cathode emission data,a small sample learning method is introduced,and the Transfer Adaboost regression algorithm is finally selected after literature research.To accommodate the algorithm,the dataset is redivided into multiple systems according to the percentage of chemical components,and the data of each system are substituted into the algorithm for prediction separately,and all of them obtain good prediction performance.Compared with machine learning regression algorithms,not only does it make full use of existing data,but the prediction effect is also improved.The results demonstrate the feasibility of machine learning as well as small-sample learning in the field of thermionic cathode research,which is a reference value for the future development of the field of thermionic cathode and the research of combining machine learning with the field of materials.