Search

Based on a Western perspective, this study utilized big data technology and web crawlers to obtain relevant ethnic and cultural information resources in Japan. Subsequently, these resources were sorted out, including the standardization and classification of content, to form a cultural resource base with national characteristics. At the same time, a BP neural network was used to evaluate the relevant factors. Finally, the accuracy and reliability of the model were verified by collecting relevant data and applying multiple linear regression methods. Among the positive probabilities, personal thoughts and social atmosphere were 0.8 and 0.66, respectively, which influenced the formation and orientation of the Japanese national cognitive structure to a certain extent. Traditional customs and religious beliefs are the factors that contribute the most to ethnicity cognition, with weights of 0.5474 and 0.4689, respectively. In the significance analysis, the p-values of dietary culture, dress culture and geographic environment are greater than 0.05, which indicates that these factors do not have a significant positive influence on ethnicity cognition.

The reliability of liquid rocket engines (LREs), which are the main propulsion device of launch vehicles, cannot be overemphasised. The development of fault detection and diagnosis (FDD) technology for LREs can effectively improve the safety and reliability of launch vehicles, which has important theoretical and engineering significance. With the rapid development of artificial intelligence technologies such as machine learning and artificial neural network, data-driven FDD methods have gained increasing attention. However, the scarcity of engine fault samples limits the application of this methods. We proposed a method combining Wasserstein generative adversarial nets (WGANs) and multilayer perceptron (MLP) to perform FDD for LREs with sample imbalance. Wasserstein generative adversarial nets were trained using the fault data from the actual hot-firing ground test of a large LRE. Considerable fault data were generated to expand the data set to balance the ratio of positive and negative samples. Subsequently, the expanded data set was used to train the MLP for FDD of a large LRE. The results showed that the samples generated by the WGAN were authentic, confirming the application of the proposed method as a novel and effective tool for establishing a complete LRE fault database. Furthermore, the diagnosis times of the proposed method on five fault tests were advanced by 0.66, 15.82, 0.24, 0.14 and 1.08 s in relation to those of the conventional red-line cut-off system. Compared with support vector machine and adaptive threshold algorithm, the proposed method also performed better. [ABSTRACT FROM AUTHOR]