CPU-GPU Cooperative Parallel Interior Point Method for Structured Nonlinear Programming

YANG Lin-feng; HU Gui-li; ZHANG Chen; ZHANG Zhen-rong

doi:10.3969/j.issn.0372-2112.2019.02.018

您当前的位置：

首页 >

文章列表页 >

CPU-GPU Cooperative Parallel Interior Point Method for Structured Nonlinear Programming

更新时间：2025-07-16

- CPU-GPU Cooperative Parallel Interior Point Method for Structured Nonlinear Programming
- Acta Electronica Sinica Vol. 47, Issue 2, Pages: 382-389(2019)
- 作者机构：
  
  1. 广西大学计算机与电子信息学院,广西,南宁,530004
  2. 广西电网有限责任公司贵港供电局,广西,贵港,537100
  3. 广西多媒体通信与网络技术重点实验室,广西,南宁,530004
  4. 广西大学计算机与电子信息学院,广西,南宁,530004
  5. 广西电网有限责任公司贵港供电局,广西,贵港,537100
  6. 广西多媒体通信与网络技术重点实验室,广西,南宁,530004
- 作者简介：
- 基金信息：
- DOI：10.3969/j.issn.0372-2112.2019.02.018
  CLC： TP301
- Published Online：25 February 2019，
  
  Published：2019
- 稿件说明：
移动端阅览
YANG Lin-feng, HU Gui-li, ZHANG Chen, et al. CPU-GPU Cooperative Parallel Interior Point Method for Structured Nonlinear Programming[J]. Acta Electronica Sinica, 2019, 47(2): 382-389.
DOI：

YANG Lin-feng, HU Gui-li, ZHANG Chen, et al. CPU-GPU Cooperative Parallel Interior Point Method for Structured Nonlinear Programming[J]. Acta Electronica Sinica, 2019, 47(2): 382-389. DOI： 10.3969/j.issn.0372-2112.2019.02.018.

摘要

大量工程应用问题可建模为结构化非线性规划，且这类问题的系数矩阵可分为稀疏型和稠密型两种类型.利用原始-对偶内点法（primal dual interior point method，PD-IPM），并结合分布式并行技术可高效求解此类问题.经典工程问题-机组组合（unit commitment，UC）为稀疏系数矩阵的结构化非线性规划，本文根据PD-IPM原理，对UC模型进行连续松弛预处理，结合快速解耦技术解耦牛顿修正方程并设计CPU-GPU协同并行算法求解子问题，最后将结果与带稠密型子问题的结构化非线性规划的求解结果进行比较和分析.实验结果显示，本文所设计的算法对于两种不同类型的结构化非线性规划求解均能获得较好的加速比.

Abstract

A lot of practical application problems can be modeled as structured nonlinearprogramming

and these problems always have two type of coefficients matrices:sparse and dense.Combining the principle of primal dual interior point method(PD-IPM) and the distributed parallel technology can solve the problem efficiently.The unit commitment(UC)is a classical engineering problem which can be formulated as a structured nonlinear programming with sparse coefficients matrices.In this paper

according to the PD-IPM principle

the UC model is continuous relaxation preprocessed

and the Newton correction equations are decoupled by using the fast decoupling technique

which can be used to obtain the independent sub problems.Then

a CPU-GPU collaborative parallel method is proposed to solve the sub problems in parallel and the results are compared with the results of structured nonlinear programming with dense sub problem.The experimental results show that the proposed method for solving two different types of structured nonlinear programming has achieved a certain speedup.

关键词

Keywords

references

Views

343

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Review on Technology of Placement and Routing for the FPGA

Build Kd-Tree for Virtual Scenes in a Fast and Optimal Way

An L₁ Exact Penalty Function Neural Network Method for Constraint Nonlinear Programming Problems

DReAC:A Novel Dynamically Reconfigurable Co-Processor

Related Author

WANG Yuan

TIAN Chun-sheng

XUE Yu

DU Zhong

MA Xiao-jing

ZHOU Chong

PANG Yong-jing

ZHANG Yao-wei

Related Institution

Beijing Microelectronics Technology Institute

Department of Micro/nanoelectronics, School of Electronics Engineering and Computer Science, Peking University

State Key Laboratory for Novel Software Technology, Nanjing University

北京应用物理与计算数学研究所第一研究室北京

北京应用物理与计算数学研究所高性能计算中心计算物理实验室北京

⁰