Scholarlink

<p>Establishing a reliable probabilistic assessment method for soil seismic liquefaction is crucial. However, the field liquefaction investigation case database used to develop robust assessment methods often contains incomplete data due to historical limitations. Traditional modeling techniques struggle to effectively utilize this incomplete information. Bayesian networks, through appropriate parameter learning algorithms, can effectively extract latent information, providing a new approach for liquefaction assessment. Five factors influencing liquefaction were selected from seismic motion conditions, environmental conditions, and soil properties. Different Bayesian network models were constructed based on expert knowledge, and EM parameter learning algorithm suitable for incomplete data conditions was applied to obtain the conditional probability table. Sample training ratio, prior weight setting, and data completeness on model overall accuracy were also explored. Additionally, forward and backward probabilistic reasoning were performed on the optimized model. Furthermore, more factors were introduced to construct 11-factors Bayesian network. The results reveal that EM algorithm is effective for parameter learning under incomplete data conditions. Model-1 is superior to other models under suitable prior weight and training ratio settings, which demonstrates more logical relationships. The accuracy was significantly enhanced under various data condition. Model-1 and the extended 11-factors Bayesian network model are recommended for soil seismic liquefaction probabilistic assessment and further study should be processed to develop suitable algorithms for incomplete data parameter learning.</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

Study on the Bayesian network probability model for soil seismic liquefaction with incomplete data

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Meng Fan,
  • Zhengquan Yang,
  • Jingjun Li,
  • Xiaosheng Liu,
  • Jianming Zhao

摘要

Establishing a reliable probabilistic assessment method for soil seismic liquefaction is crucial. However, the field liquefaction investigation case database used to develop robust assessment methods often contains incomplete data due to historical limitations. Traditional modeling techniques struggle to effectively utilize this incomplete information. Bayesian networks, through appropriate parameter learning algorithms, can effectively extract latent information, providing a new approach for liquefaction assessment. Five factors influencing liquefaction were selected from seismic motion conditions, environmental conditions, and soil properties. Different Bayesian network models were constructed based on expert knowledge, and EM parameter learning algorithm suitable for incomplete data conditions was applied to obtain the conditional probability table. Sample training ratio, prior weight setting, and data completeness on model overall accuracy were also explored. Additionally, forward and backward probabilistic reasoning were performed on the optimized model. Furthermore, more factors were introduced to construct 11-factors Bayesian network. The results reveal that EM algorithm is effective for parameter learning under incomplete data conditions. Model-1 is superior to other models under suitable prior weight and training ratio settings, which demonstrates more logical relationships. The accuracy was significantly enhanced under various data condition. Model-1 and the extended 11-factors Bayesian network model are recommended for soil seismic liquefaction probabilistic assessment and further study should be processed to develop suitable algorithms for incomplete data parameter learning.