Combined holt-winters and GA trained ANN approach for sensor validation and reconstruction: Application to water demand flowmeters

This paper proposes a Double Seasonal Holt-Winters (DSHW) forecasting model with an auxiliary Artificial Neural Network (ANN) trained with a Genetic Algorithm (GA) to model the DSHW residuals. ANN complements and improves the DSHW prediction. The proposed model also includes an on-line validation and reconstruction mechanism useful to detect and correct missing and erroneous data. This mechanism also impacts improving the DSHW prediction accuracy and precision. The proposed model and validation mechanism are applied to predict the time series generated by two monitored flowmeters of two sectors of Barcelona's drinking water network (DWN). The accuracy and precision improvement of the proposed method with respect to standard DSHW and ARIMA approaches is provided.