Soft computing techniques in the prediction of performance of semicircular breakwaters

Abstract

In the dynamic environment of the coast maintaining the harbor tranquility is possible only with the planning of proper protection structures. Breakwaters are one among the several coastal protection structures. Breakwaters could either run into the water linking to the shore or placed independently parallel to the shore. The former will lead to the accretion on up drift side and erosion on the down drift side of the structure but the latter provides shore protection without adversely affecting the longshore transport. Breakwaters attenuate the wave, slow the littoral drift and produces sediment deposition. To provide a basis for evaluating the effects of breakwater installation a comprehensive study on the hydrodynamic response of breakwaters needs to be investigated. Physical models could be used in the laboratory to assess the same however, it is expensive, laborious and time-consuming which involves many variables that affect the shape, strength, alignment, base stability and other phenomena. There are several empirical formulae but developed on limited data. Also, though numerical models are good option, it involves numerous assumptions not withstanding faster computing resources, most of which are time-consuming, tend to overestimate the hydraulic responses. The Computational Intelligence (CI) techniques can be made use to overcome some of these shortcomings. As they are capable of replicating the outcome of a numerical model with better accuracy. Among the several breakwaters available, the emerged semicircular breakwater is found advantageous and also the study on this type of breakwater is limited. Hence the present study is taken up to predict the hydraulic responses like reflection coefficient, relative wave runup, stability parameter, of emerged seaside perforated semicircular breakwater using different soft computing techniques. The soft computing techniques used are Artificial neural network (ANN), Adaptive neuro-fuzzy inference system (ANFIS), Genetic algorithm based adaptive neuro fuzzy inference system (GA-ANFIS) and Particle swarm optimization based adaptive neuro fuzzy inference system (PSO-ANFIS). The prediction is done using conventional data segregation method. Also, a methodology of segregating the lower ranges of wave height data, and not using it for training the network and then predicting the hydraulic responses purely for this segregated data is done successfully and it is namedii as ‘below the range’ predictions. Similarly, a prediction for purely higher ranges of wave height data not used in training the network, has been carried out and it is named as ‘beyond the range’ prediction. The study shows the possibility of prediction of the hydrodynamic characteristics like reflection coefficient, relative run-up parameter and stability parameter of the semicircular breakwater using the soft computing techniques for both dimensional as well as non-dimensional input parameters. In both the cases the predicted outputs the reflection coefficient, relative run-up parameter and stability parameter was good in the conventional data segregation case. Also, below the data range approach gave reasonably good results in both set of input parameters for the prediction of reflection coefficient. Whereas, in the case of beyond the data range predictions the results are good in the case of dimensional input parameters but not for non-dimensional input parameters in the prediction of reflection coefficient. The relative wave run-up parameter prediction for below and beyond the range predictions did not give satisfactory results for both set of input parameters. In the present study the stability parameter of emerged seaside perforated semicircular breakwater is predicted for a dataset of 389 data sets. The results found are good for both the set of input parameters in the case of conventional data segregation method. As the available dataset is only 389 data sets, the below the data range and beyond the data range approach was not done for stability parameter prediction. From the performance of four different models in several cases considered, the prediction made by GA-ANFIS gave better results in maximum number of cases. The ANN also predicted the output parameter well, though it is an individual model. But, the disadvantage here is the number of neurons in the hidden layer is chosen based on trial and error method, depending on thumb rules. In the case of ANFIS method the FIS could be generated by grid partitioning, subtractive clustering or fuzzy cmeans clustering. In the present study since the number of inputs in dimensional as well as nondimensional case is more than 5 the grid partitioning method has not been employed as it suffers the curse of dimensionality. In such cases the subtractive clustering or fuzzy c-means clustering can be employed. In the study it is found that the prediction made by fuzzy c-means clustering-ANFIS gave better results in maximum number of cases of reflection coefficient prediction compared to subtractive clustering-ANFIS with dimensional input parameters. Hence for all the remaining cases FCM-ANFIS is employed. The performance of PSO-ANFIS model is not as good as GA-ANFIS in the different cases considered. Arriving at the optimal parameters of the hybrid model costs time.iii However, these soft computing techniques can be adopted as an alternate technique to predict the hydraulic response of semicircular breakwaters by coastal engineers when similar site conditions are available.