Scholarlink

<p>This study proposes an improved Gaussian Process Regression (GPR) model for accurate pressure prediction within the compressor station. First, due to the limited operating conditions available from the data source of station, a steady-state compressor station simulation model is constructed for high-quality training dataset generation. Subsequently, the Latin Hypercube Sampling (LHS) is used to generate simulation data encompassing a wide range of operating conditions. Furthermore, an adaptive sampling method is employed to accurately select samples in the regions with multiple local extrema that significantly influences model accuracy. In addition, informed by physical mechanics, a hybrid kernel structure is designed to enhance nonlinear fitting ability of GPR model. The proposed GPR model is validated on both simulated and actual station data, demonstrating superior accuracy over basic data-driven models. Moreover, the data-driven approach drastically improves computational efficiency.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A High-Accuracy Gaussian Process Regression Model for Gas Flow State in Compressor Stations Based on Simulation Data

  • 期刊论文

  • 发表日期：

    2026年1月1日

查看全文
  • Weijia Li,
  • Shangfei Song,
  • Ran Liu,
  • Bohui Shi,
  • Daqian Liu,
  • Lihao Liu,
  • Weihe Huang,
  • Jing Gong

摘要

This study proposes an improved Gaussian Process Regression (GPR) model for accurate pressure prediction within the compressor station. First, due to the limited operating conditions available from the data source of station, a steady-state compressor station simulation model is constructed for high-quality training dataset generation. Subsequently, the Latin Hypercube Sampling (LHS) is used to generate simulation data encompassing a wide range of operating conditions. Furthermore, an adaptive sampling method is employed to accurately select samples in the regions with multiple local extrema that significantly influences model accuracy. In addition, informed by physical mechanics, a hybrid kernel structure is designed to enhance nonlinear fitting ability of GPR model. The proposed GPR model is validated on both simulated and actual station data, demonstrating superior accuracy over basic data-driven models. Moreover, the data-driven approach drastically improves computational efficiency.