Abstract: The comprehensive evaluation of dam service behavior is of great significance to the safe operation of water control project. Current evaluation models are too dependent on expert experience, hence lack of objectivity, or limited by data mining in the absence of professional knowledge. An assessment method for dam performance is presented based on Bayesian network combining the subjective experience of experts with the objective measured data. The relations among influencing factors are clarified effectively by determining the network structure according to naive Bayesian classification and BIC score optimization. Furthermore, the prior distribution and conditional distribution are solved by using the theory of fuzzy comprehensive evaluation and the method of entropy weight. The failure probability involving both subjective experience and objective data is obtained via reasoning through Bayesian estimation. The service behavior of dam thus can be assessed and rated according to the rating of failure probability. A multi-index evaluation model for the service behavior of dam under the Bayesian framework is established with the time series of 14 influencing factors of a concrete dam as a cases study. The safe service behavior of the dam is rated. The validity of the method is proved by analyzing the evaluation results from subjective and objective, holistic and partial, as well as static and dynamic perspectives.

Key words: dam, assessment of behavior, Bayesian network, fuzzy comprehensive evaluation, entropy weight method, failure probability