Scholarlink

<p>In urban rail flexible traction power supply system (FTPSS), conventional energy-saving strategies for reversible converter (RC) predominantly rely on offline optimization with fixed parameters. However, inherent uncertainties in train operations, such as timetable deviations and stochastic load fluctuations, result in energy consumption volatility, rendering traditional approaches suboptimal. To address this, we propose a multi-timescale model predictive control (MPC) framework that integrates day-ahead scheduling and intraday rolling optimization. Second, we propose a novel data processing method for neural network training in the intraday to construct a neural network-based load prediction model, which is used as the model prediction control (MPC) input for rolling optimization. Validated on Qingdao Metro Line 11 datasets, the prediction model achieves a correlation coefficient (<i>R</i><sup>2</sup>) value of 95.2%, and the mean squared error (MSE) is 0.078, outperforming conventional prediction methods. By integrating MPC-based rolling optimization with day-ahead scheduling, the proposed strategy improves the energy-saving rate by 2.00% over traditional offline optimization methods. Demonstrating robustness against timetable perturbations and load uncertainties.</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

Multi-timescale Optimization for Reversible Converter in DC Traction Power System Based on Model Predictive Control

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月10日

查看全文
  • Wei Liu,
  • Haonan Liu,
  • Qian Xu,
  • Juxia Ding,
  • Feilong Liu,
  • Xiaodong Zhang,
  • Dingxin Xia,
  • Haotian Deng

摘要

In urban rail flexible traction power supply system (FTPSS), conventional energy-saving strategies for reversible converter (RC) predominantly rely on offline optimization with fixed parameters. However, inherent uncertainties in train operations, such as timetable deviations and stochastic load fluctuations, result in energy consumption volatility, rendering traditional approaches suboptimal. To address this, we propose a multi-timescale model predictive control (MPC) framework that integrates day-ahead scheduling and intraday rolling optimization. Second, we propose a novel data processing method for neural network training in the intraday to construct a neural network-based load prediction model, which is used as the model prediction control (MPC) input for rolling optimization. Validated on Qingdao Metro Line 11 datasets, the prediction model achieves a correlation coefficient (R2) value of 95.2%, and the mean squared error (MSE) is 0.078, outperforming conventional prediction methods. By integrating MPC-based rolling optimization with day-ahead scheduling, the proposed strategy improves the energy-saving rate by 2.00% over traditional offline optimization methods. Demonstrating robustness against timetable perturbations and load uncertainties.