A Water Consumption Prediction Model for Municipal Consumers

Smart water resource management is the best short-time solution for water resource shortage aroundthe world. Predicting water demand is the major prerequisite to be aware of the required water within ashort time. The proposed prediction framework uses the billing records of water consumers in Yazd cityto extract consumption history. In addition, external data resources such as business calendar data, urbanwater production, meteorological parameters, the financial value of buildings, and in-stream pressureare collected and employed in the prediction model. The proposed framework tracks the changes inconsumption behaviors of consumers, which are grouped according to their volume of water usage toremove consumers with anomalous consumption behaviors. The cleaned grouped records ofconsumption are utilized in the fitting of a quantile regressor with three breakpoints to forecast the waterdemand of the consumers for the next month. Results of the experiments showed that the proposedmodel’s prediction percentage error is less than 10%. Besides, the model is able to recognize consumerswith anomalous consumption behaviors.