"Ziyue" Large Model for Yangtze River Estuary Channel Maintenance: Five Intelligent Agents Support Smart Maintenance

The Yangtze River Estuary is a world-class mega-estuary and a vital chokepoint where the Yangtze River Golden Waterway connects the river to the sea. Annual vessel cargo throughput there has long accounted for more than 50% of the total cargo volume on the Yangtze River mainline, providing important support and strong momentum for the high-quality development of the Yangtze River Economic Belt, the integrated development of the Yangtze River Delta, and the upgrading of Shanghai International Shipping Center.

The hydrodynamic and sediment conditions of the Yangtze River Estuary channel are extremely complex. To maintain the navigable depth of the 12.5-meter deepwater channel, annual dredging maintenance volume reaches as high as 50 million cubic meters, with maintenance costs exceeding 1 billion yuan. Channel depth has long depended on manual dredging, making maintenance management extremely difficult. At present, channel maintenance faces five prominent challenges: insufficient systematization of knowledge outcomes, obvious fragmentation of perception data, limited accuracy of siltation prediction and forecasting, dredging scheduling highly dependent on manual work, and a relatively low level of intelligent maintenance management.

To this end, the Yangtze River Estuary Channel Administration of the Ministry of Transport has built the "Ziyue" large model for Yangtze River Estuary channel maintenance. Relying on AI-powered digital and intelligent empowerment, it actively explores intelligent pathways for channel maintenance and has independently developed five intelligent agents covering maintenance knowledge Q&A, information perception, channel maintenance prediction, dredging vessel scheduling, and maintenance pressure assessment, building an intelligent channel maintenance system to support smart maintenance of the Yangtze River Estuary channel.

I. Tracing the Source of "Ziyue" and Carrying Forward the Sages' Aspiration to Harness the Waters

"Ziyue" is an industry large model dedicated to Yangtze River Estuary channel maintenance. The name "Ziyue" comes from the Analects: "The Master stood by a river and said: 'It passes on like this, never ceasing day or night.'" Confucius stood by the river and felt the ceaseless flow of rivers and the passage of time; today, relying on the "Ziyue" large model, we explore the laws of water and sediment movement in the Yangtze River Estuary, analyze the mechanisms of estuarine river regime evolution, and use artificial intelligence to understand the great river's changes and safeguard the channel's tranquility. Naming it "Ziyue" is precisely in the hope of carrying forward the wisdom of the sages in observing water and understanding the Way, and acting in accordance with the situation, injecting new momentum into Yangtze River Estuary channel maintenance through the power of digital intelligence.

II. Building Five Intelligent Agents to Form a Closed Loop of Channel Maintenance

First is the maintenance knowledge Q&A agent. Addressing the pain points of scattered, unsystematic knowledge in the channel maintenance field and insufficiently precise retrieval and application, the "Ziyue" large model has built a multi-domain knowledge base covering decades of achievements. Through technologies such as document vectorization and knowledge graph enhancement, it structurally processes massive reports, specifications, and research literature, enabling efficient and precise Q&A on channel maintenance knowledge. This is equivalent to equipping frontline management personnel with a portable maintenance expert, allowing them to directly query and consult on various channel maintenance-related knowledge within the system.

Second is the information perception agent. Addressing the pain points of fragmented perception data and poor reusability in channel maintenance, "Ziyue" independently built four major perception systems: monitoring, dredging, forecasting, and meteorology. On the one hand, it builds an on-site monitoring network, develops a dredging vessel perception system, develops a three-dimensional numerical model of the Yangtze River Estuary, and builds a global meteorological forecasting system; on the other hand, it carries out data labeling and governance to form high-quality datasets with broad coverage and strong perception capabilities. This agent can perceive on-site maintenance conditions in real time, equivalent to the "clairvoyance" and "clairaudience" of channel maintenance.

Third is the maintenance prediction agent. Maintenance volume prediction is the core basis for preparing channel maintenance plans. Addressing the problems of low efficiency and insufficient accuracy in traditional forecasting, especially the difficulty of sudden siltation forecasting, "Ziyue" independently developed a maintenance volume prediction algorithm that is accurate, fast-responding, and capable. Relying on the high-quality datasets and big data technologies built earlier, it independently developed machine learning models to achieve multi-scale channel maintenance volume prediction at annual, monthly, and daily levels. After model training and validation, forecast error is controlled within 10%, and response time is compressed to the second level. It has now been put into operational application and can directly guide daily channel maintenance construction arrangements.

Fourth is the dredging vessel scheduling agent. Dredging vessel scheduling is a core business process of channel maintenance. In the past, vessel dispatch relied heavily on managers' experience, resulting in low work efficiency. "Ziyue" learned from more than 3,000 historical work orders, mined vessel dispatch habits and scheduling rules, independently developed machine learning algorithms, and developed supporting algorithms for rapid identification of shallow areas and automatic generation of vessel dispatch work orders, achieving intelligent automatic scheduling of dredging vessels. It compresses the chart-reading and vessel dispatch process that originally took several hours to the second level, with a vessel dispatch plan compliance rate of 80%, significantly improving the efficiency and scientific decision-making of dredging scheduling.

Finally, the maintenance pressure assessment agent. Addressing the problems of inconsistent and insufficiently reasonable maintenance pressure assessment systems and early warning mechanisms, "Ziyue" built a maintenance pressure evaluation system containing 3 levels and 14 key indicators, and set 8 categories of early warning thresholds, covering multi-dimensional warnings such as channel depth, maintenance capacity, dumping area capacity, and daily dumping volume. Through this system, channel maintenance pressure can be dynamically assessed, enabling self-diagnosis and self-management of channel maintenance health status and early identification of various maintenance risks.

III. Strengthening Core Advantages and Promoting the Translation of Achievements into Application

Relying on 15 years of data, monitoring, and research accumulation in Yangtze River Estuary channel maintenance, "Ziyue" has built integrated real-time perception and prediction capabilities. The platform brings together five intelligent agents and integrates 10 sets of independently developed core algorithms. It has produced more than 10 patents and software copyrights, and is supported by key research and development projects, pilot tasks for building a transportation powerhouse, and multiple group standards. At present, the technical achievements have been promoted and applied, including tide level prediction for the smart channel in the Jiangsu section and construction window prediction for the second phase of the South Channel regulation project, achieving engineering translation of technical achievements.

Since being put into application at the Yangtze River Estuary, it has guided the scientific control and use of more than 1 billion yuan in dredging funds each year, and the full set of core technologies, knowledge base, and software achievements are all replicable and reusable. The research and application of the "Ziyue" large model is a vivid practice of the Yangtze River Estuary Channel Administration using digital intelligence to empower channel maintenance and safeguard the tranquility of the Yangtze River Estuary "Golden Waterway." In the future, the "Ziyue" large model will continue to be driven by scientific and technological innovation, continuously improve the capabilities of the five intelligent agents, strive to build a world-class intelligent maintenance system for estuarine channels, and contribute to accelerating the building of a transportation powerhouse and serving as a pioneer in opening the way for Chinese-style modernization.