Location:Home >> Research >> Research Progress

AI Platform for scientific research proposed by the AI Department

Date: Jul 05, 2022

  Based on the domestic heterogeneous intelligent basic software and hardware, the Department of Artificial Intelligence has proposed an artificial intelligence data and computing application service platform (AI platform") to serve multi-disciplinary cross-integration research. The AI platform meets the scientific research needs of materials computing, life sciences, finance, energy, and other disciplines. As the infrastructure of intelligent scientific computing, the platform simplifies and accelerates the iteration of AI models and data, and studies large-scale heterogeneous computing resource scheduling technology, multi-modal data fusion method, and ease of use development support environment.

  The AI platform is committed to assisting researchers and developers in different disciplines to deeply integrate data, models, algorithms, and computing resources. The platform creates an out-of-the-box interactive cloud development environment for applications in teaching, scientific research, scientific datasets, and algorithm research. Further, through the intelligent collaborative community and scientific software creative competition, the influence of the artificial intelligence platform has been expanded, and a closed ecological loop of artificial intelligence integration of multiple disciplines has been formed.

Figure1 Artificial Intelligence Science Open Platform

  Based on this platform, the Department of Artificial Intelligence and China Institute of Atomic Energy proposed a data-driven method for predicting radiation hardening behavior of RAFM steel, which solved the research problem of the synergistic mechanism of radiation embrittlement of RAFM steel. This research provides the theoretical basis and technical support for RAFM steel embrittlement mechanism and material composition optimization, and further promotes the research and development of RAFM steel for nuclear energy in my country; The Department of Artificial Intelligence and State Grid Hebei Electric Power Co., Ltd. proposed a multi-level attention mechanism prediction model combined with multi-modal domain knowledge, which solved the problem of time-series data prediction of ultra-short-term power fluctuations in centralized photovoltaics. The average forecast error was reduced by 9.5%. This research provides technical support for real-time dispatch and frequency control in smart grids; The Artificial Intelligence Department and XrayBot Artificial Intelligence Technology Co., Ltd proposed a stock price analysis method based on a knowledge graph, which solved the problem of automatic construction of a knowledge graph in the financial field and stock price analysis based on graph representation learning. This research is of great significance to industry chain analysis research and financial regulation.

Figure2 Variation of Yield Strength of RAFM Steel with Composition and Irradiation Conditions
Figure3 Prediction of new energy photovoltaic power under rainy, cloudy and sunny conditions

  Linkhttp://data.aicnic.cn/ The basic research of the AI platform has been published in the international academic journal Journal of Systems Architecture, and the applied research of the AI platform has been accepted by the international academic conference The 15th International Conference on Knowledge Science, Engineering and Management.

  For more details,please contact Wang Yangang、Wang Jue、Cao Rongqiang(wangyg@sccas.cn、wangjue@sccas.cn、cncaorq@sccas.cn)

Appendix: