The Research And Implementation Of Bayesian Dynamic Modelling For Oceanographic Time Series Forecasting

21st century is an ocean generation, the major issue of which is sustainabledevelopment,therefore, the strategic and development value of marine hasincreasingly become the focus of all the country in the world. The coastal economyaccounts for a large proportion of global GDP,and the ration has increased year byyear.Growing faster than GNP growth, faster than any other industries.While thedevelopment of China’s marine economy has entered the orbit of world economy, andmaine economy has developed as new growth economy in the most countries of theworld.However, with the rapid economic development, the formation of a largegroup of coastal cities, the concentration of both population and property caused byurbanization,global warming and rising sea levels, making marine disasters occurredfrequently. Losses to China’s coastal areas is increasing, and gradually made Chinabecome one of the worst marine disasters suffered countries in the world.If we can make timely and more accurate dynamic prediction, and make ascientific and effective decision in the shortest period of time,then we can reducedisaster losses to some extent, which becomes an urgent task of disaster preventionand mitigation for marine research.Hydrological forecasting is one of a more directway, which has some significance for disaster prevention and mitigation. It is used tomake a qualitative and quantitative prediction for the future state of a region, a bodyof water and a hydrological station, basing on information which is known.This is animportant aspect for economic and social services,which is of great significance forflood, drought, rational use of water resources, national economy and even nationaldefense.The main currently ways for hydrological forecasting methods are stepwiseregression, multiple linear regression, autoregressive, wavelet analysis, artificialneural networks, time series forecasting techniques and so on. Time series forecastingmethod is used in this paper., it analyses time series variables based on the use ofmathematical methods to establish predictive model, so that the series trend extendsoutwards to obtain the development trends of time series and get the predictive variables’ values. Nonstationarity, caused by climate forcing and other factors, such aschange in physical properties of catchment(urbanization, vegetation change, etc.),makes the forecasting task too difficult to model by traditional Box–Jenkinsapproaches.However, the basic assumption of all such models is the temporalpersistence of statistical properties of the time series.This is often not a validassumption.Another drawback of Box–Jenkins models is the exploitation of asignificant amount of data for determining the parameters and for validation themodel before it becomes ready for use.Large data sets may not always be available,particularly in developing countries.Our study is based on the project named Bohai Sea storm surge disasterassessment and decision support system, our aim is the water depth values oftwo-dimensional flood evolution.The main research content of this article can bedivided into the following parts:Firstly,parameter extraction for the prediction models.100water depth values ofa flood evolution is used here, using MATLAB tool to draw out the time-seriesimages, and95%confidence level in the autocorrelation curve. We can get the initialforecasting conditions through the analysis of values, the experience of forecastersand experts. While predicting, the observed values are added, which fully reflects thedynamic of Bayesian models. After100times of the cycle,we obtain the posteriordistribution, and the predicted water depth value.Secondly,analysis of the predicted results. Through the contrast betweenprediction and observation value curves and the prediction error curve, the Bayesiandynamic model is worked well here, the average relative error of only5.73%, whichis well in line with forecast peak error of20%. Operational needs can be received, canbe extrapolated to the industry. The model is also used recursive approach to calculatethe cycle, and therefore especially suitable for online real-time monitoring, which forthe flood period, snow, storm surges and other disasters caused by floods during thereal-time monitoring is very meaningful. It also shows the dynamic model andBayesian prediction method is suitable for marine hydrological time series such asnon-stationary time series. This is the prediction of hydrology, especially in themodern climate change under extraordinary circumstances, has a certain significance.Finally,implementation of the predicted results. Evolution of the flood waterdepth of100times the value of the data to predict, the evolution of the two-dimensional flood prediction applied to the Bohai Sea storm surge disasterassessment and response decision support system for two-dimensional flood routingsubsystem. By the user to select a simulation, combined with a series of intermediateserver-side logic processing, call stored in the relational database back-end forecastresults, and then the front-end interface, you can view the time evolution of the floodsimulation. Development and use of this system Flex, Java as the primarydevelopment language, back-end database to adopt a more mainstream databaseORACLE. The system uses an advanced three-tier architecture (client application/browser (B/S architecture) application server/middleware, database) to facilitatesystem management and maintenance, can improve system security.