Prediction For Water Quality Of Dissolved Oxygen Based On Time-series Data

The wide application of Internet of Things technology,makes the amounts of water environment monitoring data growing exponentially.How to mine and analyse the mass dynamic water environment monitoring data,to find the knowledge and rules,and to predict the concentration changes of the index of water quality,are of great significance on the management and decision making of hydrological environment.Dissolved oxygen(DO)is one of the important water quality parameters that reflects the situation of water pollution.The measurement and monitoring of concentration changes of DO in the water environment is vital for water quality management.Based on the research status analysis of water quality prediction method,and on the DO time-series data of Yuancuo station and its two adjacent stations of Minjiang river in Fuzhou,we carried out DO prediction research respectively based on multi ways,such as,on the different number of variables with single variable time series and multivariate time series,and on different number of models with single prediction model and the multi combined model.The specific research work and conclusions include the following aspects:(1)We preprocessed the outliers,missing value of water quality time-series data respectively by boxplot and cubic spline interpolation method,then probed and analyzed the historical water quality time-series data as follows:we analyzed the mean value of DO in four seasons of Yuancuo station from 2009 to 2014 and got seasonal law of DO value,that is,"high in winter low in summer";based on the wavelet analytic method,we found that time-series data of DO and water temperature both have law of cycle for 12 months;we visualized DO and water temperature time-series data from 2011 to 2013 based on the calendar map and found DO is negatively related to the water temperature,the reason of DO that "high in winter low in summer" is likely to be affected by the seasonal impact of water temperature:that is,DO will be low in summer when the water temperature is high,and high in winter when the water temperature is low.(2)Trough the theory study of time-series prediction method,grey model,and artificial neural network,we established single water quality prediction model respectively based on three kinds of prediction method,that is,time-series,gray model,artificial neural network.Then we predicted the DO value from the 1st to the 6st week of 2015 according to DO time-series data of Yuancuo site from the 1st week of 2013 to the 52nd week of 2014.(3)We built the Grey Neural Network Prediction Model based on DO time-series data of Yuancuo site in Fuzhou,and built the Principal Component Neural Network Prediction Model based on the DO time-series data of Yuancuo site in Fuzhou and its two adjacent site from 2013 to 2014,and then predicted the DO value from the 1st to the 6st week of 2015 respectively.Finally,we compared the prediction effect of DO value from the 1st to the 6st week of 2015 of Yuancuo site in Fuzhou based on the above five kinds of prediction methods within the certain prediction step lengh.The results showed that the comprehensive prediction effect for the sampel data of the study area based on the multivariate time-series and the Principal Component Neural Network Prediction Model is excellent within the five models;and the prediction effect based on the Grey Neural Network Prediction Model with single variable of the sampel data is better than the single model of grey model and the neural network prediction model.The conclusion can provide a reference for the choice of prediction model of DO in the study area.