Prediction Of Seawater Index Based On Phase Space Reconstruction And Bidirectional Gated Recurrent Unit Neural Network

With the exploitation of marine resources,the pollution of marine environment is becoming more and more serious.To know the status of seawater quality in time and reduce the loss caused by seawater pollution,it is of great significance to accurately forecast seawater index.Therefore,in this study,the phase space reconstruction(PSR)and Bi GRU are used as the research methods in the sea area of Beihai City,Guangxi Zhuang Autonomous Region.Based on the data of seawater temperature,p H,dissolved oxygen and salinity collected by the buoy of ocean monitoring station,the prediction model is established.The results of this study provide key technical support for the seawater quality big data analysis platform of the sea area,and play an important role in understanding the marine conditions of the sea area.The main research contents of this paper are as follows:Firstly,the data collected by the buoy is preprocessed,including outlier processing and data partitioning.The autocorrelation analysis and chaos characteristic analysis of the preprocessed data are carried out.The univariate prediction model of seawater index based on PSR-Bi GRU is established,and the four seawater indexes of multiple monitoring points are predicted respectively.It is compared with the more excellent chaotic time series prediction model at present.In this way,the validity of the model is verified.Compared with the comparison model,the prediction model of PSR-Bi GRU has the best prediction effect in predicting seawater index.Secondly,the marine environment is complex and changeable,and noise is unavoidable in seawater data.The existence of noise can reduce the prediction effect of the model.To solve this problem,a noise reduction and feature extraction method based on improved CEEMDAN(ICEEMDAN)is proposed to reduce the noise interference to the final prediction results.The ICEEMDAN-PSR-Bi GRU univariate prediction model of seawater index is established.Four sea water indexes at multiple monitoring sites were taken as prediction objects,and compared with the PSR-GRU model,EMD-PSR-Bi GRU model and CEEMDAN-PSR-Bi GRU model,respectively,so as to understand the influence of different noise reduction methods on the prediction results.And the effectiveness of the noise reduction method is verified.Compared with the other three models,RMSE,MAE and MAPE of ICEEMDAN-PSR-Bi GRU model are all decreased,while R~2 is increased.Finally,the prediction of seawater index is affected not only by the prediction variable itself,but also by other indexes.Therefore,considering the correlation between each index and establishing a multivariate prediction model of seawater index can enhance the interpretability of the prediction model and improve the prediction effect.The complex correlation coefficient is used to analyze the correlation among the four indexes,and all indexes are selected as the input of the prediction model according to the analysis results.At the same time,multivariate phase space reconstruction is carried out for the four indexes.The PSR-Attention-Bi GRU(PSR-Att Bi GRU)multivariable prediction model of seawater index is established.The model enhances the capturing ability of data information by adding Attention.The model is compared with univariate PSR-Bi GRU model and multivariate PSR-Bi GRU model.The results show that the multivariate prediction model can improve the prediction effect to a certain extent,and the addition of Attention also has a positive effect on the improvement of the prediction effect.