Predicting Domestic Water Consumption in Urban Areas Using Smart Water Meter Data

Rapid urbanization and population increase in Kenyan cities have caused increased water demand and severe challenges in managing water consumption. Additionally, water leakage and variable human activities produce non-linear patterns that challenge the predictive accuracy of traditional linear time series models, such as the autoregressive integrated moving average (ARIMA). To address this limitation, this study introduces a Long Short-Term Memory (LSTM) network model tailored to predict water demand. Using Nairobi City as a case study, the model employs high-frequency non-linear time series data collected from smart meters between January and December 2022. Nine variables were constructed from the data to train and validate the LSTM network, which was then evaluated using root mean square error (RMSE) and the correlation coefficient. The results reveal significant daily and monthly variations in domestic water consumption, underscoring the complexity of urban water demand patterns. Despite these variations, the LSTM model outperformed the ARIMA model, generating predictions that closely aligned with observed values while achieving lower prediction errors. The LSTM model's ability to capture non-linear dynamics highlights its suitability for urban water consumption forecasting. The findings suggest that this model could be generalized to predict future water consumption in urban households across Kenya and similar regions. Additionally, the study demonstrates the value of smart meter data in enabling more accurate, data-driven approaches to managing water resources in urban environments.