Enhancing multi-component alloy composition prediction based on generative adversarial networks and proximal policy optimization.

This article belongs to the cross fusion of material genomics engineering and artificial intelligence. In response to the challenges of data scarcity and high experimental costs faced by traditional alloy design methods in complex composition systems, an innovative intelligent algorithm combining generative adversarial networks (GAN) and proximal policy optimization (PPO) is proposed, and a new research paradigm of "data generation, intelligent optimization, experimental verification" is constructed. This method uses the GAN module to generate high-quality alloy samples of tens of thousands of levels with only a hundred level initial experimental data, effectively alleviating the problem of data scarcity. At the same time, the PPO algorithm is used to transform alloy composition design into a Markov decision process, which significantly improves search efficiency in high-dimensional combinatorial space through dynamic interaction and optimization between intelligent agents and the environment. Compared with traditional optimization algorithms, this method demonstrates significant advantages in computational efficiency, data utilization, and component prediction accuracy. It can significantly reduce experimental costs and shorten development cycles, providing new ideas for the design of high-performance multi-component alloys. This study not only provides scalable intelligent algorithm tools for material genome engineering, but also offers a new research method for reverse design of complex material systems, which has important scientific significance and application value.