Scholarlink

<p>The satellite clock bias (SCB) time series exhibits nonlinear and non-stationary characteristics, with interference between components potentially affecting prediction accuracy. Therefore, We propose a SCB prediction model based on singular spectrum analysis (SSA), integrated with an adaptive neuro-fuzzy inference system (ANFIS). Specifically, we initiate the process by employing SSA to decompose and reconstruct the first-order difference sequence of clock bias, thereby extracting the dominant component and the residual component. Subsequently, the ANFIS model is employed to forecast the reconstructed components, and the predicted outcomes are overlaid and reinstated to derive the ultimate predicted clock bias value. Lastly, through experiments, the proposed model is compared with the grey model (GM (1,1)), quadratic polynomial model (QP), long short-term memory neural network model (LSTM), and ANFIS model. We chose all BDS-3 satellites, and the findings indicated that the traditional QP model and GM (1,1) model were ineffective in prediction. However, the SSA-ANFIS model can significantly enhance the prediction accuracy compared to individual models. In comparison to the ANFIS model, its 12&#xa0;h prediction accuracy increased by 45.60%. Similarly, when compared to the LSTM model, the prediction accuracy rose by 32.04%.</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

Highly accurate short-term prediction of BDS-3 satellite clock bias based on SSA-ANFIS integrated modeling

  • 期刊论文

  • 发表日期：

    2025年4月30日

查看全文
  • Kaihui Lv,
  • Chenglin Cai,
  • Yihao Cai,
  • Wenhui Guan,
  • Yun Fang

摘要

The satellite clock bias (SCB) time series exhibits nonlinear and non-stationary characteristics, with interference between components potentially affecting prediction accuracy. Therefore, We propose a SCB prediction model based on singular spectrum analysis (SSA), integrated with an adaptive neuro-fuzzy inference system (ANFIS). Specifically, we initiate the process by employing SSA to decompose and reconstruct the first-order difference sequence of clock bias, thereby extracting the dominant component and the residual component. Subsequently, the ANFIS model is employed to forecast the reconstructed components, and the predicted outcomes are overlaid and reinstated to derive the ultimate predicted clock bias value. Lastly, through experiments, the proposed model is compared with the grey model (GM (1,1)), quadratic polynomial model (QP), long short-term memory neural network model (LSTM), and ANFIS model. We chose all BDS-3 satellites, and the findings indicated that the traditional QP model and GM (1,1) model were ineffective in prediction. However, the SSA-ANFIS model can significantly enhance the prediction accuracy compared to individual models. In comparison to the ANFIS model, its 12 h prediction accuracy increased by 45.60%. Similarly, when compared to the LSTM model, the prediction accuracy rose by 32.04%.