The Application Of Data Pre-processing And Histogram Time Series In Water Quality Prediction

With the rapid economic development in China, the shortage, waste and pollution of water resources have become increasingly prominent problems. Water quality early warning system can effectively control and reduce the harm caused by the deterioration of water quality through real-time water quality monitoring, forecasting, assessment and early warning. Finally we can achieve the target of effective cognition and timely control of the deterioration of water quality.In this thesis, the research of water quality prediction was conducted on the basis of the review of literature. There are some problems in water quality data collected by the current water quality warning system, such as data redundancy, data missing, etc. While traditional prediction methods pay few attentions to the quality of water quality data. The water quality data observe in high frequency must be predicted and analysed in low frequency. Traditional prediction method uses the means of the data as the sample data. A lot of information is neglected. Regarding the current issues, this thesis has carried out related research work. The main results and innovation points are summarized as follows:1. A series of data pre-processing methods, based on k-means clustering, LIN interpolation method and single spectrum analysis, are proposed to improve the quality of water quality data. The BP neural network model is used to verify the effectiveness of the data pre-processing methods. The results show that it can improve the reliability and accuracy of the water quality prediction.2. The histogram time series is introduced to deal with the high frequency water quality data collected by the water quality warning system. In this thesis, a new method is proposed to construct histogram time series data which is a kind of symbolic data.3. The histogram time series generated from the high-frequency raw data is used to predict the water quality. We use KNN model and exponential smoothing model in histogram time seires analysis for predicition. This thesis carried out a comparative study between KNN model, smoothing model and naive model.The data pre-processing methods and histogram time series prediction methods proposed in this thesis can improve the reliability and accuracy of the water quality prediction. The histogram time series prediction can also provide more distribution information than traditional water quality prediction method.