Application Of Neural Network Based On Genetic Algorithm On Wave Impact Force Prediction

In coast and offshore engineering, the safety of the structures whose superstructures located in the splash zone such as piled wharves, shore trestles, oil drilling platforms, etc,have great relation to wave slamming. Under confused sea condition, the structures may subject to very strong wave impulsive load due to slamming by the wave with significant crest when waves propagate underneath the structure and surge up to its subface. Previous studies indicate that impact pressures are characterized by an initial peak pressure of considerable magnitude but of short duration occurs, followed by a slowly-varying uplift pressure of less magnitude but of considerable duration, and which typically is first positive, then decreases to zero and becomes negative.In hostile sea state, the peak pressures may cause the damage of the horizontal members of the structures or make the whole superstructure collapsed. Also the concrete structure's service life may be threatened by the negative pressure occurred when waves separate frome the subface of the structures. In the recent twenty yesrs, the wave impact force prediction has become the important research. However, so many effective factors in the wave impact force as well as the complicated changing, non-linear, therefore the difficulties come into being to the research of the wave impact force prediction.In the past few years, with the development of information technology, especially the Artificial Intelligence Technology, a new way has been come out for wave impact force prediction. In this dissertation, the most advanced Neural Network technology in Artificial Intelligence was introduced to the field of the wave impact force prediction, and with the direction of this new idea, the research has been done in the wave impact force prediction based on the Artificial Neural Network.According to the disadvantage of the traditional BP algorithm, such as the local minimum points and the slow learning speed, the improve method of Momentum combined with Variable Learning Rate Back Propagation (BPX), Conjugate Gradient Back Propagation (CGFBP), Levenberg Marquardt Back Propagation (LMBP), the first kind of the LMBP combined with Genetic Algorithm (GABP1) and the second kind of the LMBP combined with Genetic Algorithms (GABP2) are choosen. These improved algorithms are applied to train the net. The end, the optimum algorithm (GABP1) and parameters of the net and the algorithm, such as the number of the hidden units, aim error, weights and bias, selection operator, crossover operator,mutation operator, crossover probability, mutation probability, population size and so on are received. According to the relationship between the wave impact force and wave height, wave period and the relative clearance of underside of the wharf deck and so on, a wave impact force predicton model based on the Genetic Neural Network is designed and developed. The prediction results have shown that the better accuracy of wave impact force prediction can be achieved by the model applied the GABP1 for modeling.