Performance Optimization Model of Molecular Dynamics Simulation Based on Machine Learning and Data Mining Algorithm

Abstract

Molecular dynamics simulation is an effective method for related research in the microscopic world, mainly related to molecular atomic research, and it has a wide range of applications in physics, medicine, and other disciplines. Meanwhile, with the expansion of data to be studied, the performance of molecular dynamics simulation in terms of calculation speed and other aspects can no longer fully meet the needs of today’s data analysis and calculation. On the basis of the original method, parallel calculation and optimization improvement of the method have become the focus of researchers. This paper will use machine learning (ML) and data mining algorithms to optimize the performance of molecular dynamics simulation and use Restricted Boltzmann Machines (RBM) and K-Nearest Neighbors (KNN) in the molecular dynamics simulation system. Through the parallel optimization experiment of KNN, the effectiveness of KNN optimization is obtained, and finally the molecular dynamics simulation optimization experiment is designed. Through comparative analysis with the simulation system before optimization, it is concluded that when the number of particles is 4096, the efficiency ratio of force calculation running time is the highest of 31.15%. When the number of particles is 512, the running time efficiency ratio of the motion trajectory equation is up to 30.28%. When the number of particles is 256, the efficiency ratio of running time tending to balance judgment is the highest of 36.96%. All the results show that the performance of the optimized simulation system has been improved. The experimental results are in line with expectations.