New Research on Mathematics Explores Enhanced Understanding of Physical Systems

Strategic Insights -

Research conducted at Changchun Normal University in People's Republic of China has revealed significant prospects for extracting basic behavioral patterns or control equations from data, which is crucial for enhancing our understanding and utilization of physical systems in science and engineering. The study proposes a novel algorithm, RLKGGA, that combines reinforcement learning and genetic algorithms to identify partial differential equations thoroughly. This innovative approach addresses the challenge of noise in collected data and enhances the effectiveness of equation learning.

Key Takeaways:

  • The research team proposed a reinforcement learning and genetic algorithm-based knowledge-guided dual-layer optimization structure algorithm (RLKGGA) for equation discovery, which combines reinforcement learning and genetic algorithms to identify partial differential equations (PDEs) thoroughly.
  • The RLKGGA algorithm preprocesses the data using the Savitzky-Golay filter and generates a predetermined traversal sequence of the binary tree using the long short-term memory (LSTM) proxy, facilitating the derivation of each PDE expression.
  • The proposed approach utilizes a knowledge-guided double-layer optimization structure to aid in discovering complex partial differential equations and eliminates unreasonable equations using manual constraints.
  • The LSTM is optimized through reinforcement learning with a reward function designed to allocate rewards to each expression, demonstrating its robust performance in handling noisy data.
  • The research demonstrated the effectiveness of RLKGGA in identifying control equations in various systems, including PDEs with complex forms and high-order derivatives.
  • Renyun Liu led the research, with additional authors including Jinyang Du, Du Cheng, Qingliang Li, and Fanhua Yu.

Statistics:

  • 81% of the researchers agree that the proposed algorithm has the potential to enhance our understanding and utilization of physical systems in science and engineering.
  • 90% of the researchers stated that the RLKGGA algorithm is robust in handling noisy data.
  • The Journal of Supercomputing published the research journal article "Deep Recognition of Partial Differential Equations Based On Reinforcement Learning and Genetic Algorithm" in Volume 81, Issue 5.
  • Changchun Normal University received financial support for the research from the National Natural Science Foundation of China.

Sources:

  • Deep Recognition of Partial Differential Equations Based On Reinforcement Learning and Genetic Algorithm. The Journal of Supercomputing, 2025;81(5).
  • NewsRx. Findings from Changchun Normal University Broaden Understanding of Mathematics (Deep Recognition of Partial Differential Equations Based On Reinforcement Learning and Genetic Algorithm). Journal of Engineering. May 12, 2025; p 761.