Due to an increase in the demand for water and the limited availability of real time monitoring technologies, efficient management of residential water resources poses a unique problem. This study describes the development of an inexpensive Internet of Things (IoT)-based electronic system capable of providing real time monitoring and artificial intelligence (AI)-driven analysis of residential water consumption. The electronic system consists of the use of a YF-DN50 flow sensor connected to an ESP32 microcontroller allowing for continuous data acquisition. The data collected is stored in a cloud-based spreadsheet on Google Colab for processing and analysis, where an unsupervised Long Short-Term Memory (LSTM) Autoencoder model was developed for anomaly detection. The developed model produced a final training loss value of 0.038 and a validation loss value of 0.045, with the threshold for anomaly detection determined to be mean + 3 standard deviations above the reconstruction error. During the observation period, the system detected a total of 7 anomalous events as well as detecting an overall 85.7% of the detected known events and having a 33% reduction in false negative detections compared to a simple fixed threshold baseline. The results of this study demonstrate that the combination of IoT sensors with LSTM Autoencoder analysis of water consumption allows for the monitoring of water consumption patterns and early detection of anomalies, thus providing a means to achieve a reduction in average monthly household water consumption from approximately 21 m3 to 18 m3.
Quinto et al. (Wed,) studied this question.