| Abstract |
Water quality monitoring is essential for ensuring the safety and sustainability of water resources and protecting public health and the environment. However, water quality data may contain anomalies, which are deviations from the normal behavior of the data. Anomalies can be caused by various factors, such as sensor faults, environmental disturbances, human interventions, or malicious attacks. Anomalies can affect the accuracy and reliability of water quality assessment and management and may lead to false alarms, missed detections, or incorrect decisions. Therefore, it is important to detect and remove anomalies from water quality data in a timely and effective manner. In this paper, we propose a novel multivariate deep learning technique, called Hybrid Multivariate Long Short-Term Memory (HM-LSTM), for detecting anomalies in water quality monitoring using multivariate time series data. HM-LSTM is a hybrid model of multiple performing neural networks and long short-term memory networks that can effectively learn and detect anomalies from water quality data. We apply our technique to a real-world water quality dataset collected from industrial fields in the Middle East and compare it with several baseline methods. We show that our technique can achieve higher performance and provide detailed information about the water status and the types and causes of anomalies. We also provide explanations for the anomaly detection results by using the attention mechanism and the anomaly score. Our technique can benefit from the spatial and temporal features of the data, and enhance the anomaly detection performance by focusing on the most relevant features. Our technique can be applied to other domains and scenarios that involve multivariate time series data, such as smart cities, smart health, smart agriculture, and smart industry Copyright 2024, Society of Petroleum Engineers. |
