Adaptively weighted support vector regression: Prognostic application to a historic masonry fort

Prognostic evaluation involves constructing a prediction model based on available measurements to forecast the health state of an engineering system. One particular prognostic technique, support vector regression, has had successful applications because of its ability to compromise between fitting accuracy and model complexity in training prediction models. In civil engineering applications, compromise between fitting accuracy and model complexity depends primarily on the measured response of the system to loads other than those that are of interest for prognostic evaluation, referred to as extraneous noise in this paper. To achieve accurate prognostic evaluation in the presence of such extraneous noise, this paper presents an approach for optimally weighing fitting accuracy and complexity of a support vector regression model in an iterative manner as new measurements become available. The proposed approach is demonstrated in prognostic evaluation of the structural condition of a historic masonry coastal fortification, Fort Sumter located in Charleston, South Carolina, considering differential settlement of supports. Within this case study, the adaptive optimal weighting approach had increased forecasting accuracy over the nonweighted option.