Scholarlink

<p>The positioning of the FAST feed cabin measurement system relies on GPS/IMU fusion technology. However, in GPS-denied environments, the system degrades to pure IMU mode. In this case, errors accumulate during the operation of the feed cabin and result in significant positioning deviations. To address the challenges of maintaining GPS/IMU fusion in GPS-denied environments, this paper develops a hybrid model integrating a long short-term memory (LSTM) network with multi-model adaptive estimation (MMAE). The LSTM effectively mitigates IMU error accumulation through its temporal dependency modeling, achieving 67% higher computational efficiency compared to traditional IMU processing methods that rely on complex numerical integration and Kalman filtering algorithms. The MMAE adaptively blends optimized IMU data with theoretical trajectories through mode-based adaptive weighting. In the primary operating modes (accounting for 67% of total tasks), the model achieved positioning errors below 0.04&#xa0;ms in 84% of tasks. For another subset of operating modes (15% of total tasks), due to most tasks exceeding one hour in duration, long tasks were divided into multiple independent 20-minute segments to improve positioning accuracy, successfully reducing positioning errors to below 0.05&#xa0;ms in these scenarios. The experimental results demonstrate that during GPS outages, the model can effectively suppress drift caused by standalone IMU operation over time, providing relatively accurate positioning, and can serve as a viable alternative solution in GPS-denied environments.</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

Enhancing Positioning Accuracy of FAST Feed Cabin in GPS-Denied Environments: A LSTM-MMAE Fusion Model for IMU Drift Suppression

  • 期刊论文

  • 发表日期：

    2025年12月26日

查看全文
  • Yu Feng,
  • Minghui Li,
  • Mingjie Chen,
  • Benning Song,
  • Dongjun Yu

摘要

The positioning of the FAST feed cabin measurement system relies on GPS/IMU fusion technology. However, in GPS-denied environments, the system degrades to pure IMU mode. In this case, errors accumulate during the operation of the feed cabin and result in significant positioning deviations. To address the challenges of maintaining GPS/IMU fusion in GPS-denied environments, this paper develops a hybrid model integrating a long short-term memory (LSTM) network with multi-model adaptive estimation (MMAE). The LSTM effectively mitigates IMU error accumulation through its temporal dependency modeling, achieving 67% higher computational efficiency compared to traditional IMU processing methods that rely on complex numerical integration and Kalman filtering algorithms. The MMAE adaptively blends optimized IMU data with theoretical trajectories through mode-based adaptive weighting. In the primary operating modes (accounting for 67% of total tasks), the model achieved positioning errors below 0.04 ms in 84% of tasks. For another subset of operating modes (15% of total tasks), due to most tasks exceeding one hour in duration, long tasks were divided into multiple independent 20-minute segments to improve positioning accuracy, successfully reducing positioning errors to below 0.05 ms in these scenarios. The experimental results demonstrate that during GPS outages, the model can effectively suppress drift caused by standalone IMU operation over time, providing relatively accurate positioning, and can serve as a viable alternative solution in GPS-denied environments.