AquaBrain: Self-Adaptive Optimization Framework for Water Quality Prediction
Technology Transfer & Commercialisation

AquaBrain: Self-Adaptive Optimization Framework for Water Quality Prediction

Silver Medal

AquaBrain: Self-Adaptive Optimization Framework for Water Quality Prediction

Water pollution has become a serious problem affecting human health around the world. Timely detecting upcoming water pollution can help make treatment plans in advance. Recently, deep learning techniques are introduced to predict future water quality. Yet existing methods customized the intelligent model architectures and water features for certain water areas by experts with trial-and-error schemes and are manually re-designed when applied to other scenarios as water environments are varied. The invention proposes a self-adaptive optimization framework for water quality prediction, consisting of an extreme water event detector, a water quality prediction model, and a heuristic-based optimization controller. Except analyzing the data from the time-series viewpoint, the framework can track the changes in water indicators from an event-based viewpoint to discover unusual water events. More importantly, the controller enables the system to automatically adjust the architecture and hyper-parameters of prediction models, and the choice of appropriate water quality indicators for forecasting. Therefore, the system can quickly adapt to various water areas without input knowledge from AI experts and water experts while reducing manpower costs and still maintaining outstanding performance.