Scholarlink

<p>Near-Earth Objects (NEOs) are scientifically important because they pose potential hazards to Earth and provide insights into Solar System dynamics. Accurate classification and orbital prediction are essential for scientific research and planetary defence. The conventional approach tends to address the two problems independently and separately and is too reliant on noisy or partial observations, which limits their applicability for long-term prediction and real-time monitoring. One limitation is the lack of integrated models capable of grouping NEOs into statistically derived behavioural clusters, and the second problem is to make predictions of their near-future close approaches to the planet. To tackle this issue, a hybrid artificial intelligence model was created that uses both unsupervised and supervised learning for efficient NEO analysis. Using MiniBatchKMeans clustering on the data between 1910 and 2024, the model categorises NEOs into clusters of similar physical and orbital features like size, speed, and encounter distance. These cluster labels are subsequently utilised as contextual features in a time-series forecasting pipeline that combines Long Short-Term Memory (LSTM) networks with Extreme Gradient Boosting (XGBoost) regression. The LSTM picks up on patterns in orbital movement over time, and the XGBoost minimises residuals to achieve a maximum level of prediction accuracy. The Principal Component Analysis (PCA) and robust feature scaling were used to ensure robustness in noise processing and dimensionality reduction. The scalability, and interpretability of the model, enables scalable analysis of large observational datasets in a research context, and visualisations like PCA plots and prediction comparison can be validated by domain experts. The hybrid model is an exploratory method of analysis of NEO monitoring data based on data. The given framework can be viewed as a screening and prioritization tool that relies on data instead of physics-based orbital propagation. This paper uses the term cluster to mean statistically determined groupings in feature space, and has nothing to do with classical collisional asteroid families in proper orbital-element space.</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

Artificial intelligence-driven clustering and prediction of Near-Earth Object behavioural groups and close approaches

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • Dolly Sharma,
  • Misha Shah,
  • Aviichal Sharma,
  • Mamta Mittal

摘要

Near-Earth Objects (NEOs) are scientifically important because they pose potential hazards to Earth and provide insights into Solar System dynamics. Accurate classification and orbital prediction are essential for scientific research and planetary defence. The conventional approach tends to address the two problems independently and separately and is too reliant on noisy or partial observations, which limits their applicability for long-term prediction and real-time monitoring. One limitation is the lack of integrated models capable of grouping NEOs into statistically derived behavioural clusters, and the second problem is to make predictions of their near-future close approaches to the planet. To tackle this issue, a hybrid artificial intelligence model was created that uses both unsupervised and supervised learning for efficient NEO analysis. Using MiniBatchKMeans clustering on the data between 1910 and 2024, the model categorises NEOs into clusters of similar physical and orbital features like size, speed, and encounter distance. These cluster labels are subsequently utilised as contextual features in a time-series forecasting pipeline that combines Long Short-Term Memory (LSTM) networks with Extreme Gradient Boosting (XGBoost) regression. The LSTM picks up on patterns in orbital movement over time, and the XGBoost minimises residuals to achieve a maximum level of prediction accuracy. The Principal Component Analysis (PCA) and robust feature scaling were used to ensure robustness in noise processing and dimensionality reduction. The scalability, and interpretability of the model, enables scalable analysis of large observational datasets in a research context, and visualisations like PCA plots and prediction comparison can be validated by domain experts. The hybrid model is an exploratory method of analysis of NEO monitoring data based on data. The given framework can be viewed as a screening and prioritization tool that relies on data instead of physics-based orbital propagation. This paper uses the term cluster to mean statistically determined groupings in feature space, and has nothing to do with classical collisional asteroid families in proper orbital-element space.