Multi-Task Learning-Based Integrated Water Quality Assessment and Pollution Source Tracing Model

• Published in: 2024 4th International Conference on Artificial Intelligence, Robotics, and Communication (ICAIRC) pp. 998 - 1001
• Main Authors: Tao, Sheng, Liu, Haiyang, Wang, Hongshuo, Men, Yingxin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 27.12.2024
• Subjects: Accuracy; Monitoring; Multi-task Learning; Multitasking; Neural networks; Pollution; Pollution Source; Predictive models; Robustness; Tracking Neural Networks; Turbidity; Water pollution; Water quality; Water Quality Assessment
• DOI: 10.1109/ICAIRC64177.2024.10899971

In light of the mounting issue of water pollution, the precision and real-time capability of water quality monitoring have emerged as a pivotal subject in the domain of environmental conservation and public health. This paper presents a comprehensive water quality assessment and pollution source tracking model based on multi-task learning (MTL). On the basis of the existing water quality prediction and pollution source identification models, we propose a multi-task learning framework which jointly trains multiple related tasks. This framework is designed to enable the collaborative optimisation of water quality parameter assessment and pollution source tracking. In particular, we extend the traditional single-task model to jointly learn water quality indicators and pollution source characteristics through shared representations, and simultaneously carry out pollution source location prediction within the same framework. The model employs a multi-branch neural network architecture, wherein each branch is tasked with a distinct objective but interacts with information through a shared network layer, thereby facilitating knowledge transfer and enhancing the model's generalisation ability. Furthermore, a dynamic weighting mechanism is introduced to ensure a balanced contribution of each task and enhance the robustness of the model. The experimental results demonstrate that the proposed model exhibits superior performance in terms of water quality prediction accuracy and pollution source tracking efficiency compared to existing advanced technology.