Research On Short-term Prediction Of Sea Surface Temperature, Salinity And Wind Speed Based On Neural Network Metho

The accurate prediction of marine hydrometeorological data is very important for human activities such as developing marine resources,preventing marine disasters and protecting ecological environment.With the increasing of offshore operations and marine production activities,the timely and accurate prediction of sea surface temperature,sea surface salinity and sea surface wind speed are becoming more and more critical.Neural network is the key research object in the field of prediction.This thesis takes the neural network prediction model as the research method,carries on the short-term prediction based on the marine data monitored by the marine observation buoy,and explores a reliable research route of data prediction in the field of marine science.The main research contents of this thesis are as follows:Firstly,the accurate short-term prediction of sea surface temperature and salinity data is carried out.When the prediction days are fixed at 5 days,the mean square error of prediction results of different training days is compared,and it is determined that the prediction results are relatively optimal with the observation data of 20 days as the training set.Then,taking the sea surface temperature and sea surface salt data of January,April,July and October 2009 obtained by single site observation buoy as the training set,BP and RBF neural network models are trained respectively and applied to the prediction of sea surface temperature and sea surface salt data on the 21 st to 25 th days of each month.The results show that both BP and RBF neural network can effectively deal with the seasonal changes of data,and the overall prediction effect of RBF neural network is better than BP neural network.Finally,through the prediction experiment of multisite data,it is verified that the RBF neural network prediction model has strong applicability and accuracy.Secondly,a short-term sea surface wind speed prediction model based on LSTM neural network is proposed.The historical wind speed data collected by a single buoy station in Ocean SITES database is selected as the input and the prediction model is trained.Then,the short-term prediction experiment of sea surface wind speed is carried out by setting different prediction time,and the prediction results are compared with those of BP and RBF neural network prediction models,which proves the superiority of LSTM neural network prediction method.Finally,the universal applicability of LSTM neural network model is verified by the prediction experiment of station data corresponding to multiple sea areas.From the comparative analysis of the prediction error and the correlation coefficient verification between the predicted value and the real value,it can be seen that this method still has the stability of prediction when dealing with the rapidly changing data.Finally,the self-developed multi parameter marine monitoring buoy is used to carry out marine experiments in relevant sea areas,and the practicability of the neural network prediction model is predicted based on the received marine parameters.Through the prediction of sea surface temperature,salt and wind speed,the effectiveness of neural network model for short-term prediction is proved.By comparing the prediction results of the experiment,it can be seen that the prediction error of LSTM neural network model for data with gentle change trend is slightly higher than that of RBF and BP neural network models,but the overall performance is still good.Through the analysis of prediction error and correlation coefficient between predicted value and real value,it is proved that LSTM neural network can predict all kinds of marine data with high precision and high prediction stability.Similarly,RBF neural network model also has reliability when dealing with marine data with relatively flat change.