Multi-model prediction for demand forecast in water distribution networks

This paper presents a multi-model predictor called Qualitative Multi-Model Predictor Plus (QMMP+) for demand forecast in water distribution networks. QMMP+ is based on the decomposition of the quantitative and qualitative information of the time-series. The quantitative component (i.e., the daily co...

Descripción completa

Detalles Bibliográficos
Autores: López Farías, Rodrigo, Rodríguez-Rangel, Héctor, Puig, Vicenç, Flores, Juan J.
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2018
País:España
Institución:Consejo Superior de Investigaciones Científicas (CSIC)
Repositorio:DIGITAL.CSIC. Repositorio Institucional del CSIC
OAI Identifier:oai:digital.csic.es:10261/179261
Acceso en línea:http://hdl.handle.net/10261/179261
Access Level:acceso abierto
Palabra clave:Prediction
Multi-model
Water demand
Short-term prediction
Descripción
Sumario:This paper presents a multi-model predictor called Qualitative Multi-Model Predictor Plus (QMMP+) for demand forecast in water distribution networks. QMMP+ is based on the decomposition of the quantitative and qualitative information of the time-series. The quantitative component (i.e., the daily consumption prediction) is forecasted and the pattern mode estimated using a Nearest Neighbor (NN) classifier and a Calendar. The patterns are updated via a simple Moving Average scheme. The NN classifier and the Calendar are executed simultaneously every period and the most suited model for prediction is selected using a probabilistic approach. The proposed solution for water demand forecast is compared against Radial Basis Function Artificial Neural Networks (RBF-ANN), the statistical Autoregressive Integrated Moving Average (ARIMA), and Double Seasonal Holt-Winters (DSHW) approaches, providing the best results when applied to real demand of the Barcelona Water Distribution Network. QMMP+ has demonstrated that the special modelling treatment of water consumption patterns improves the forecasting accuracy