陆军工程大学学报2024，Vol.3Issue(4) ：60-67.DOI:10.12018/j.issn.2097-0730.20240318005

机器学习优化地理空间域计算强度预测与分解方法

A Machine Learning Optimized Approach for Computational Intensity Prediction and Domain Decomposition of Geospatial Domain

高凡 李娜 甘麟露
陆军工程大学学报2024，Vol.3Issue(4) ：60-67.DOI:10.12018/j.issn.2097-0730.20240318005
✕
  • NETL
  • NSTL
  • 万方数据

机器学习优化地理空间域计算强度预测与分解方法

A Machine Learning Optimized Approach for Computational Intensity Prediction and Domain Decomposition of Geospatial Domain

高凡 1李娜 2甘麟露1
扫码查看

作者信息

  • 1. 陆军工程大学通信工程学院,江苏南京 210007
  • 2. 61175部队,江苏南京 210049
  • 折叠

摘要

并行计算被广泛应用于地理空间大数据分析,其中利用地理空间域表征计算强度实现负载均衡是目前并行地理计算中的热点.针对现有地理空间计算强度建模方法过分依赖专家知识,导致建模过程复杂、适用性差的问题,从数据科学的角度出发,提出了一种利用机器学习预测地理空间域计算强度的方法.以矢量空间相交为例,分析提取矢量相交的候选特征空间,结合机器学习特征选择和回归拟合算法,自动寻找最优地理空间域特征空间,从而实现地理空间域计算强度的准确预测,在此基础上完成地理空间域的均衡分解.实验对比了机器学习优化方法和传统域分解方法的性能,结果证明了所提方法的可行性和高效性.该方法为运用机器学习构建地理空间域计算强度预测模型和优化地理空间域分解提供了参考.

Abstract

The parallel computing has been widely used in the analysis of geospatial big data,and a-mong them,to represent computational intensity with geospatial domain to achieve load balancing is cur-rently a hot spot in parallel geocomputing.However,the existing geospatial computational intensity mod-elling methods are overly dependent on expert knowledge,resulting in a complex modelling process and poor applicability.In order to address this problem,this paper proposes a method for predicting the com-putational intensity of geospatial domain based on machine learning from the perspective of data science.Taking the intersection of vector spaces as an example,this paper analyzes and extracts candidate feature spaces for vector intersection,combines machine learning feature selection and regression fitting algo-rithms,automatically searches for the optimal feature space in the geospatial domain,thereby achieving accurate prediction of computational intensity of the geospatial domain.Based on this,a balanced decom-position of the geospatial domain is accomplished.The performance of the machine learning optimization method is compared with that of the traditional domain decomposition method by experiments,and the re-sults demonstrate the feasibility and efficiency of the proposed method.This method provides a reference for using machine learning to build prediction models of computational intensity of the geospatial domain and optimize the decomposition of the geospatial domain.

关键词

并行计算/地理空间域/计算强度/机器学习

Key words

parallel computing/geospatial domain/computational intensity/machine learning

引用本文复制引用

基金项目

国家自然科学基金青年项目(42301489)

江苏省自然科学基金青年项目(BK20231030)

出版年

2024
陆军工程大学学报
解放军理工大学科研部

陆军工程大学学报

影响因子：0.556
ISSN：2097-0730
段落导航相关论文