Scholarlink

<p>Accurate prediction of particulate matter with an aerodynamic diameter smaller than 2.5&#xa0;μm (PM<sub>2.5</sub>) is considered to be essential for effective air pollution warnings to be issued and health impacts to be mitigated. This study presents a novel predictive model tailored to PM<sub>2.5</sub> concentration time series that are characterized by high noise levels, strong volatility and non-stationarity. The proposed method integrates variational mode decomposition (VMD) with a multi-kernel hybrid relevance vector machine (HRVM) optimized by an improved sparrow search algorithm (ISSA), referred to as VMD-ISSA-HRVM. First, VMD decomposes the non-stationary PM<sub>2.5</sub> sequence into a set of stationary modal components. Next, a sparse multi-kernel HRVM is constructed to forecast each modal component individually, with ISSA optimizing the weight coefficients of the hybrid kernel function. The predictions of all sub-models are then combined to produce the final forecast. Four RVM-based forecasting approaches were evaluated using real-world PM<sub>2.5</sub> data from Beijing. The results demonstrate that the VMD-ISSA-HRVM model achieves the highest prediction accuracy and generalization capability, outperforming conventional RVM models by improving simulation accuracy by a factor of 104 and enhancing generalization accuracy by approximately 94%.</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

PM2.5 prediction using hybrid relevance vector machine with ISSA-optimized multi-kernel and variational mode decomposition

  • 期刊论文

  • 发表日期：

    2025年6月10日

查看全文
  • Lin Xu,
  • Jia-Hao Zhang,
  • Chun-Wu Yin

摘要

Accurate prediction of particulate matter with an aerodynamic diameter smaller than 2.5 μm (PM2.5) is considered to be essential for effective air pollution warnings to be issued and health impacts to be mitigated. This study presents a novel predictive model tailored to PM2.5 concentration time series that are characterized by high noise levels, strong volatility and non-stationarity. The proposed method integrates variational mode decomposition (VMD) with a multi-kernel hybrid relevance vector machine (HRVM) optimized by an improved sparrow search algorithm (ISSA), referred to as VMD-ISSA-HRVM. First, VMD decomposes the non-stationary PM2.5 sequence into a set of stationary modal components. Next, a sparse multi-kernel HRVM is constructed to forecast each modal component individually, with ISSA optimizing the weight coefficients of the hybrid kernel function. The predictions of all sub-models are then combined to produce the final forecast. Four RVM-based forecasting approaches were evaluated using real-world PM2.5 data from Beijing. The results demonstrate that the VMD-ISSA-HRVM model achieves the highest prediction accuracy and generalization capability, outperforming conventional RVM models by improving simulation accuracy by a factor of 104 and enhancing generalization accuracy by approximately 94%.