Modular Tidal Level Forecasting Based On BP Neural Networks

Tidal level forecasting is significant for shipping transportation, port construction and tidal energy development. Along with the development of shipping business and the request for navigation safety and efficiency of shipping, higher requirements are put forward for the accuracy of tidal level prediction. The application of neural networks in tidal level forecasting became an important research area in recent years. Back propagation (BP) neural network has been widely in pattern recognition and system forecast field. Tidal level prediction modules combined with BP neural network are proposed in this thesis and the applications of BP neural network in tidal level prediction are discussed.The conventional method for tidal level forecasting is the harmonic analysis method which only considers the effect of celestial bodies to tidal level, therefore the prediction accuracy decreased significantly in areas with complex environmental factors. In view of this situation, a direct tidal level prediction module using BP neural network only is designed. Based on the measured tidal level data,the accuracy of short-term tidal level prediction is improved.Modularization design is a method for solving nonlinear problems. A modular tidal level prediction model based on BP neural network is proposed by detailed analyzing the constituent of tide. The modular model contains the harmonious analysis module for predicting time-varying portion causing by celestial bodies and BP neural network module for predicting the residual portion causing by other elements. The reliability of modular model is improved by achieving distinction of prediction function. By combining the advantages of harmonic analysis method which can provide stable, long-term astronomical tidal level prediction and BP neural network which can achieve the prediction of other parts of tidal level, prediction accuracy is improved further.The prediction accuracies of harmonic analysis method, BP neural network method and modular model are compared by simulation. The result shows that for short-term tidal level forecasting, the performance of modular model is better than harmonic method and BP method.