Author(s): Null
Linked Author(s):
Keywords: River and lake health; Deep learning; Nested model; Self-iteration; Boundary condition optimization
Abstract: The current paradigm of river and lake condition assessment is often constrained by the limitations of singular indicators, the absence of real-time feedback in evaluation outcomes, and the incomplete information synthesis in expert-driven decision-making processes. This study introduces a novel AI model architecture characterized by its nested structure, designed to overcome these challenges. The model leverages a two-tiered approach: the outer layer is trained on a comprehensive datasets of health indicators for rivers and lakes, establishing the boundary conditions that delineate the health status of water bodies. The inner layer is tailored to individual river or lake, ingesting all the collected data to refine the outer layer’s parameters and introduce additional boundary conditions. The methodology involves vectorization and normalization to generate input samples for a deep learning network model, which enables an iterative training regimen within the nested model framework, enhancing the model’s predictive accuracy and adaptability over time. The case study focuses on city-level rivers and lakes within Yunnan Province, utilizing health assessment reports and “one river/lake, one policy” documents as primary data sources. Empirical findings indicate that the AI model is proficient in identifying and diagnosing the health issues pertaining to specific river or lake. It could analyse underlying causes of these issues and provide prescriptive measures, auxiliary by supporting horizontal references. This approach offers a significant advantage over traditional policy reports by reducing the number of proposed engineering interventions and associated financial outlays.
Year: 2024