News

Accurate utility demand forecasting with use of smartphone location data

18-03-2020

Location data harvested from a mobile phone are a valuable source of information with many potential applications related to planning and management. Such application - a water demand prediction model - has been presented by a group of scientists from Institute of Geodesy and Geoinformatics and Institute of Environmental Engineering from Wrocław University of Environmental and Life Sciences and University of Auckland, New Zealand. The study applied over seven millions of smartphone location data combined with historical water usage data. Datasets were integrated into the machine learning model, creating an accurate water demand forecasting model.



Accurate water demand forecasts can be a useful tool for efficient water supply system management. The knowledge about real water demand is a crucial element for water pressure adjustment. Demand underestimation leads to low water pressure, limiting ist accessibility. On the other hand, overestimation of water demand is far much dangerous, causing high system pressure, which in results increases the risk of water leaks, which is directly related to significant losses. Therefore, reasonable water pressure adjustment leads to substantial savings by limiting water leaks as well as lowering energy consumption by water pumps. Moreover, accurate water demand forecasts serve as an early-response system for potential major networks faults, helping to find leaks, when measured water usage is much higher than predicted.

According to recently published research, applying mobile phones’ location data improves the accuracy of short-term water demand forecasts. The study was conducted on five different areas with different water usage characteristics - residential, industrial and mixed - forecasting next 7 days and 24 hours of water usage. At the current stage of development, the location data improves models predictions accuracy by 1%, which is a good result in comparison to other studies employing other sources of data, like for example temperature and rainfall (2,5% of improvement). Average predictions accuracy was 90,5%.

Applying mobile phones’ location data to water demand prediction task is a unique approach, never published before. The good results are a great starting point for further studies. Specifically, the published article studies the possibility of forecasting water demand using only location data, that is excluding historical consumption data. This solution may be interesting for water supply managers who cannot afford to install expensive water flow measurement devices on their network, giving them a low-cost solution for efficient water supplying system management.

The published research article can be found here:
https://doi.org/10.1080/1573062X.2020.1734947.

Data for this study was made available local water supply infrastructure management company MPWiK and Selectivv Mobile House.