Water inrush is one of the most serious geological disasters threatening tunnel construction. Generally, complexity and multi-sources feature of physical information existing in tunnel construction make disaster prediction very difficult, how to accurately predict the disaster becomes a hot topic in the field of tunnel engineering. Dempster-Shafer (DS) theory of evidence is a widely used method for reasoning with multiple evidences, however some unbelievable results usually appear in dealing with highly conflicting evidences by its traditional combination rule. Thus an improved fusion algorithm based on weighted average of evidence conflict probability was firstly introduced into risk prediction of water inrush disaster. Through the improved algorithm, multi-sources precursor information measured from previous model test were fused to predict quantitative risk levels of water inrush for different excavation step of subsea tunnel in the model test. The predicted high risk at the 12th excavation step by improved algorithm agreed well with actual phenomenon of intensive seepage observed in the test, while the traditional method gave a lower level. Moreover, the improved algorithm predicted a more accuracy result in the phase of water inrush (at 16th excavation step shown in test). In brief, the improved algorithm can make more accuracy prediction for water inrush disasters and will provide valuable reference for similar engineering.

突水是威胁隧道建设的最严重的地质灾害之一。通常，隧道施工中存在的物理信息的复杂性和多源性特征使得灾害预测非常困难，如何准确预测灾害成为隧道工程领域的热门话题。Dempster-Shafer（DS）证据理论是一种用于对多个证据进行推理的广泛使用的方法，但是在通过其传统组合规则处理高度冲突的证据时，通常会出现一些令人难以置信的结果。因此，首先将基于加权平均证据冲突概率的融合算法引入到突水灾害的风险预测中。通过改进的算法，融合先前模型测试中测得的多源先驱信息，以预测模型测试中海底隧道不同开挖步骤的突水定量风险水平。改进算法预测的第十二个开挖步骤的高风险与试验中观察到的强烈渗漏的实际现象非常吻合，而传统方法给出的水平较低。此外，改进的算法预测了突水阶段（测试中所示的第16个挖掘步骤）的准确性更高。简而言之，改进后的算法可以更加准确地预测突水灾害，为类似工程提供有价值的参考。改进算法预测的第十二个开挖步骤的高风险与试验中观察到的强烈渗漏的实际现象非常吻合，而传统方法给出的水平较低。此外，改进的算法预测了突水阶段（测试中所示的第16个挖掘步骤）的准确性更高。简而言之，改进后的算法可以更加准确地预测突水灾害，为类似工程提供有价值的参考。改进算法预测的第十二个开挖步骤的高风险与试验中观察到的强烈渗漏的实际现象非常吻合，而传统方法给出的水平较低。此外，改进的算法预测了突水阶段（测试中所示的第16个挖掘步骤）的准确性更高。简而言之，改进后的算法可以更加准确地预测突水灾害，为类似工程提供有价值的参考。