Abgeschlossene Arbeiten seit 2004
Nutzen Sie die Recherchemöglichkieten im Forschungsinformationssystem (FIS) zur weiteren Einschränkung der aufgelisteten Abschlussarbeiten.
Error detection and correction as pre-processing steps of high resoulution water quality time series.
Art der Abschlussarbeit
Master
Autoren
- Ngo, Hai Nam
Betreuer
- Dipl.-Hydrol. Björn Helm
- Dipl.-Hydrol. Stefanie Wiek
- Prof. Dr. sc. techn. Peter Krebs
Abstract
Missing values often occur in time series data monitoring pollutants of specific water bodies and water technical systems. To overcome the lack of these data points and provide more complete analysis of the studied data, missing imputation is used to provide the highest accuracy of the missing data. Imputation uses statistical methods to estimate the possible missing values based on the internal relationship within the time series. This work includes four methods of imputation from simple, less accurate to complex, considerably higher accuracy; specifically Average Imputation, Linear Interpolation, Autoregressive Integrated Moving Average (ARIMA) and Mulitvariate Imputation. Water quality data of rivers in Saxony, Germany were used to verify the methods and perform filling of actual missing values. Calculation was done by R-language coding and executed in Studio R, an open source language. The results show the accuracy of data increasing with the sophistication of imputation methods used. ARIMA demonstrates the lowest error or a missing data gar lager than 15 points when compared to other univariate imputation methods. High resolution time series imputation error is significantly smaller than aggregate time series data (hourly, daily). Multivariate imputation works well when the test data is missing more than 50% of the total data points. The error introduced by multivariate imputation increases slightly with the size of the missing gap or aggregated time series. Automatic imputation methods are chosen and applied to an entire time series based on the missing gap size of actual data.
Schlagwörter
Imputation, water quality, pre-processing
Berichtsjahr
2014