Machine learning methods for alarm detection and prediction

点击翻译按钮获取中文摘要

Water quality monitoring is becoming increasingly important. The importance of tracking changes in water quality can't be overstated: human health and livelihoods depend on clean, reliable water supplies. Remote real time water quality monitoring systems specializing in drinking water, surface water, ground water, industrial water and waste water are becoming more feasible and economical than ever. These systems are some of the first to integrate Internet of Things (IoT) technologies and remote sensing oriented to health and safety. As with all advances, being early adopters in integrating sensors, systems and software on scale presents considerable research questions and challenges. Of particular concern in this Engage Grant opportunity, is how machine learning can be best leveraged to help predict alarms from real time water quality data while mitigating against false alarms. The data is inherently temporal and presents challenges that interpreting spatial data does not involve. Although machine learning has made astonishing progress in image recognition and in learning strategies or policies in games such as chess or go, time series data is still of the most difficult to analyze. This research project will evaluate machine learning alternatives and ensemble methodolgies (such as those based on support vector machines, convolutional neural networks, recurrent neural networks as well as more traditional neural networks) with the objective to determine the most suitable machine learning approach to improve the prediction of water quality alarm states as well as the mitigation of false positives.********