Abstract: In order to objectively obtain accurate and reliable rock quality evaluation results, a heterogeneous ensemble evaluation method based on improved D-S (Dempster-Shafer) evidence theory is proposed for the existing individual evaluation models which are prone to misjudgment and bias. Firstly, based on determining the rock mass quality evaluation standard and data set, 6 heterogeneous base classifiers are used to train the data set, and the preliminary rock mass quality evaluation results are obtained. Then, the result of each base classifier is taken as a piece of evidence, and the evidence is integrated through the improved D-S evidence theory to form a comprehensive evaluation result, to solve the inconsistency of the results of the existing evaluation methods. Based on 30 groups of samples, a heterogeneous ensemble evaluation model based on improved D-S evidence theory was constructed and applied to the rock mass quality evaluation of the Guangzhou water pumping and storage power station phase II project. The results show that the heterogeneous ensemble evaluation method based on the improved D-S evidence theory improves the accuracy and reliability of rock mass quality grade identification, avoids subjectivity and uncertainty in the evaluation process, and has high engineering practical value.

Key words: rock mass, rock quality evaluation, D-S evidence theory, ensemble learning, heterogeneous ensemble