Research On Seawater Quality Prediction Method Based On Relevance Vector Machine Theory

Seawater,as the carrier of the entire ocean ecosystem,plays an important part in the construction of ocean ecological civilization.The changes in seawater quality not only directly affect the diversity of ocean life,but also destroy the stability of ocean ecosystem causing ocean disasters.Seawater quality forecasting and early warning can effectively prevent water quality from deterioration.It is of great significance to ocean ecological and environmental protection.On the basis of the data from a certain sea area in China,the relevance vector machine(RVM)theory is introduced for predicting and evaluating the seawater quality.Main studies are focused on data preprocessing methods related to seawater quality,prediction and evaluation model and parameter optimization methods.The main details are as follows:1)In order to solve the problem that the numerous data of seawater quality collected by sensors are correlated with each other and their dynamics are difficult to analyze,a dimensionality reduction extraction method based on principal component analysis is proposed for seawater quality data.By calculating the principal component loading factor matrix,the influence of each water quality factor on the principal component is analyzed.The simulation results show that this method not only realizes data dimensionality reduction and key influence factor extraction,but also reduces computational complexity.2)In order to solve the problem of low prediction accuracy of single-kernel RVM,an adaptive multi-kernel RVM prediction model is proposed.The model consists of three parts:data dimensionality reduction,nonlinear approximation and parameter optimization.The data dimensionality reduction extraction provides a high-quality sample set for the predictor.The RVM uses combined kernel functions to approximate the seawater quality data nonlinearly.The firefly algorithm is used to optimize the RVM parameters.The simulation results indicate that,compared with the single-kernel RVM model,the model has better nonlinear approximation ability.3)In order to solve the problem of multiple coupling factors in water quality evaluation and inaccurate determination of water quality grade by a single factor,a water quality evaluation model derived from fuzzy comprehensive evaluation and RVM is proposed.The fuzzy comprehensive evaluation method is used for the quantitative calculation of multiple indicators to determine the water quality level.The RVM with different kernel functions is used for classification.The simulation results demonstrate that,compared with linear kernel and polynomial kernel,the RVM based classification model with Gaussian kernel has a higher classification accuracy.4)In order to solve the problem that the model building process is computationally complex and non-computer professionals are not familiar with model building,a seawater quality prediction and early warning system is designed.The system software mainly includes five functional modules: data query,data visualization,seawater quality data prediction,statistical analysis of predicted results,and seawater quality data evaluation and early warning.Through many experiments on the system,it is proved that the designed system can effectively monitor seawater quality.Figure 24;Table 10;Reference 64...