The purpose of this paper is to present a simple yet highly effective method to reconstruct missing data in flow time series. The presence of missing values in network flow data severely restricts their use for an adequate management of billing systems and for network operation. Despite significant technology improvements, missing values are frequent due to metering, data acquisition and storage issues. The proposed method is based on a weighted function for forecast and backcast obtained from existing time series models that accommodate multiple seasonality. A comprehensive set of tests were run to demonstrate the effectiveness of this new method and results indicated that a model for flow data reconstruction should incorporate daily and seasonal components for more accurate predictions, the window size used for forecast and backcast should range between 1 and 4 weeks, and the use of two disjoint training sets to generate flow predictions is more robust to detect anomalous events than other existing methods. Results obtained for flow data reconstruction provide evidence of the effectiveness of the proposed approach.
Data reconstruction of flow time series in water distribution systems – a new method that accommodates multiple seasonality
Rui Barrela, Conceição Amado, Dália Loureiro, Aisha Mamade; Data reconstruction of flow time series in water distribution systems – a new method that accommodates multiple seasonality. Journal of Hydroinformatics 1 March 2017; 19 (2): 238–250. doi: https://doi.org/10.2166/hydro.2016.192
Download citation file: