Nature-inspired metaheuristic optimization in least squares support vector regression for obtaining bridge scour information

The scouring of stream and river channels is a complicated phenomenon; it is a function of flow energy, sediment transport, and bridge substructure characteristics that challenges bridge engineers worldwide. Scour is also a major cause of bridge failure, thus contributing substantially to the total construction and maintenance costs of a typical bridge. Accurately estimating local scour depth near bridge piers is vital in engineering design and management. Thus, an effective technique is necessary to estimate the safety and economy of bridge design and management projects. This study developed a novel hybrid smart artificial firefly colony algorithm (SAFCA)-based support vector regression (SAFCAS) model for predicting bridge scour depth near piers. The SAFCAS integrates a firefly algorithm (FA), chaotic maps, adaptive inertia weight, Lévy flight, and support vector regression (SVR). First, adaptive approaches and randomization methods were incorporated into the FA to construct a novel metaheuristic algorithm for global optimization. An SVR model was then optimized through SAFCA to maximize its generalization performance. Laboratory and field data reported in the literature were applied to evaluate the proposed hybrid model. The effectiveness of the proposed intelligence fusion system was evaluated by comparing the SAFCAS modeling results with those of numerical predictive models and with the results of empirical methods. For the bridge scour depth problem, the proposed hybrid model achieved 3.99%-87.12% better error rates compared with other predictive methods, as measured through cross-fold validation algorithms and hypothesis testing. The resulting SAFCAS model can infer decisive information to assist civil engineers in designing safe and cost-effective bridge substructures.