面向氢储存的金属氢化物生成式机器学习设计模型

Developing novel metal hydrides is a critical step towards efficient hydrogen storage in carbon-neutral energy systems. However, existing materials databases, such as the Materials Project, contain a limited number of well-characterized hydrides, which constrains the discovery of optimal candidates. This work presents a framework that integrates causal discovery with a lightweight generative machine learning model to generate novel metal hydride candidates. Initially, causal discovery techniques are employed to analyze existing hydride data, identifying key structural and chemical features that influence hydrogen storage performance. These features are then encoded as input parameters for the model, capturing the intrinsic relationships between material properties and functionalities. Subsequently, a deep learning-based generative model is trained to learn and mimic the structure-property relationships of known hydrides. The core of this model is a variant of a Variational Autoencoder (VAE) or Generative Adversarial Network (GAN), designed to generate new material structures with specific desired attributes from a latent space. Through iterative optimization, the model can explore the vast chemical space of materials design, generating potential metal hydrides with high hydrogen storage capacity, excellent cycling stability, and suitable thermodynamic properties. The advantage of this framework lies in its ability to transcend the limitations of existing databases, actively exploring unknown material combinations. The generated candidate materials can then be validated through first-principles calculations or experimental synthesis to confirm their predicted performance. This approach is expected to accelerate the high-throughput material screening process, significantly shortening the research and development cycle for next-generation efficient hydrogen storage materials.