Neural network for determining the characteristic points of the bars

This article focuses on the optimal architecture of the neural network for determining the three characteristic points of the bars (starting, crest and final point). For the definition of the network, precision profiles, sedimentological and wave data were used. A total of 209 profiles taken for 22 years was used. The inputs were analysed and selected considering the variables that influenced the formation of the bars and their movement. For the selection of the optimal model different architectures were studied, generating 50 models for each of them and selecting with better results and with the smaller number of neurons in the hidden layer. To evaluate the performance of the model, various statistical errors were used (absolute error, mean magnitude of relative error and percentage relative error), with an average absolute error of 17.3 m in the distances to the coast and 0.26 m in the depths. The results were compared with equations currently employed (Table 1), which show that the errors generated by the ANN (Artificial Neural Network) are much lower (per example the MAPE committed by the proposed equation for distance to shore of the crest is 47%, while the ANN is made of 29%).This research has been partially funded by Universidad de Alicante through the project “Estudio sobre el perfil de equilibrio y la profundidad de cierre en playas de arena” (GRE15-02)