Sewer Structural Condition Prediction Integrating Bayesian Model Averaging with Logistic Regression

Utility managers and other authorities often rely on sewer structural condition prediction models for the effective execution of long-term and short-term sewer management strategies; however, it is challenging to predict the structural condition effectively because of the intrinsic uncertainties in modeling. In this research, a Bayesian framework is developed to predict the structural condition of sewers considering model uncertainties. Bayesian model averaging (BMA) techniques are used for identifying significant covariates for different sewers considering model uncertainties, whereas Bayesian logistic regression models are applied for predicting the structural condition of sewers. To validate the effectiveness of the proposed framework, the structural condition of 12,728 sewer mains of the wastewater network of the city of Calgary, Canada, is predicted. The results show that the BMA approach provides a transparent statement of the posterior probabilities to represent the effect of the significant explanatory covariates, and the performance of the Bayesian logistic regression model improves with informative priors.