Deep Decision Tree Generation And Optimization Technology Based On Aerospace Big Data

In the aerospace field,the use of traditional physical models to study complex turbulence problems has always had a variety of limitations.With the advent of the era of artificial intelligence,machine learning methods provide a new way for us to solve this type of problem.Based on the decision tree model,this paper explores the effects of classical decision tree,parallel decision tree and deep decision tree in constructing a new subgrid-scale model,and finally improves the deep forest model(gcForest)to obtain a new deep decision tree model with superior performance and interpretability-DF2022.The main features of the new model are as follows:1.Suitable for non-microdata model building.The model is modeled after a deep neural network design,using a cascading forest structure instead of the backpropagation of the neural network,thus providing a possibility for building a high-performance non-micro deep learning model.2.Optimize the memory footprint of the traditional deep forest model.The model mainly optimizes the memory occupation of gcForest from three aspects:feature histogram,sample subsampling and complementary feature compression.3.Further improve the ability of the model to predict.The model mainly improves the prediction ability of the gcForest model from two aspects:customizing multi-granular scanning rules according to expert knowledge and increasing the complexity of cascading forests layer by layer.4.The model is interpretable.The model is composed of a forest model arranged hierarchically,from the routing rules of each subtree in the forest to multi-granular scanning and data transfer between cascading layers,all of which have good interpretability.