| In view of frequent water pollution events,this paper based on data-driven model to carry out time and space prediction and early warning analysis of water quality,to provide a basis for the decision and disposal of sudden water pollution events.For the prediction model of time series of water quality,the variational mode decomposition algorithm is used to decompose the time series of water quality into independent and stable sub-series.Then,Pearson correlation analysis was used to determine the correlation between different water quality indexes and each subsequence,which was used as the input variable of BP neural network.The BP neural network water quality prediction model based on variational modal decomposition was established,and the chlorophyll concentration in the case was predicted and analyzed.Compared with the BP neural network model based on empirical mode decomposition(EMD)and the BP neural network model for water quality prediction,it is found that variational mode decomposition can avoid the problem of EMD mode mixing,and can effectively reduce the non-stationarity of water quality time series.Pearson correlation analysis can judge the correlation between water quality and subsequence and optimize the input of neural network.The R~2of the established model is 0.95,which has a high accuracy and can provide a basis for time series prediction and early warning.For the prediction model of water quality spatial migration,based on the information entropy theory,the fast correlation filter algorithm is used to screen the multi-dimensional upstream hydrological water quality variables related to the downstream target water quality on the spatial and temporal scale according to the relationship between the upstream hydrological water quality and the downstream water quality,which are used as the input of BP neural network.The prediction model of water quality spatial migration based on BP neural network based on fast correlation filter algorithm was established,and the nitrate concentration and electrical conductivity in the case were predicted and analyzed.Compared with the single BP neural network model,R~2increased by 0.1.The fast correlation filter algorithm can effectively filter the input of BP neural network and optimize the model structure.The model can effectively reflect the migration and change of pollutants in the river and provide effective information for spatial water quality warning.Aiming at sudden pollution events,abnormal detection of water quality was carried out from two perspectives of time and space,and semi-supervised water quality time and space warning models based on support vector machine were established respectively.By using the accuracy of the model established by the area judgment under the receiver operating characteristic curve and comparing one-class support vector machine and support vector data description under different kernel functions.It was found that the accuracy of support vector description method was 85%.Finally,qualitative analysis and comprehensive weighted index method are used to evaluate the warning grade of water quality time series warning,and to judge the danger area of water quality spatial migration warning model.In this paper,a prediction and early warning model of water quality is established from the perspectives of time and space,which can effectively reflect the changes of water quality in time and space,and carry out an accurate early warning analysis.It provides decision-making information for environmental managers and technical support for the establishment of water quality prediction and early warning model,and the established model has good generalization. |
