The research was led by Jianwei Zhang, Weitao Wu, and Ge Hou of the School of Water Conservancy at North China University of Water Resources and Electric Power in Zhengzhou, together with Jinlin Huang of the Guangdong Research Institute of Water Resources and Hydropower in Guangzhou and Biao Jia of the Inner Mongolia Water Conservancy Science Research Institute in Hohhot. Their starting point was the structural character of embankments, the earthen barriers that form the backbone of many reservoirs. Rather than relying on a single measurement, the team constructed a comprehensive evaluation index system that captures three distinct streams of evidence: continuous monitoring data from instruments, the findings of on-site inspections, and review indicators drawn from formal safety assessments.

Once the indicators were defined, the researchers confronted a classic dilemma in multi-criteria evaluation: how much should each factor count? Subjective weighting methods, which rely on expert opinion, capture engineering experience but can introduce personal bias. Objective weighting methods, which are derived mathematically from the data itself, avoid bias but may ignore knowledge that experienced inspectors bring. The team calculated weights separately using both approaches and then fused them through an improved game theory framework built on the idea of minimizing the overall loss of information. In game theory terms, the subjective and objective weight vectors are treated as players whose combined outcome should preserve as much information as possible from both sources, yielding fusion weights that neither approach could achieve alone.

With the weights settled, the model had to handle the deep uncertainty that pervades dam monitoring. Instrument readings drift, inspection findings are qualitative, and the boundary between safe and unsafe is fuzzy rather than sharp. To convert between these qualitative judgments and quantitative numbers, the researchers employed forward and reverse cloud generators, the computational engines of the cloud model, a cognitive framework first proposed by Li Deyi and colleagues in 2009. A forward cloud generator transforms linguistic concepts, such as a rating of normal or abnormal, into populations of cloud droplets, which are random numerical realizations that scatter around the concept according to defined statistical properties. A reverse cloud generator performs the opposite conversion, distilling raw data back into the numerical characteristics of a concept.

The cloud model’s dependence on randomly generated cloud droplets introduces its own subtle error: the number of droplets used in a calculation changes the resulting membership degrees. The team tackled this by developing a calculation method based on membership degree that quantifies the error generated by different numbers of cloud droplets, allowing a reasonable optimization of the droplet count. This optimization matters because too few droplets produce unstable, noisy results, while too many waste computational effort without meaningfully improving accuracy. By calibrating the droplet number against the error it induces, the model achieves a balance that keeps its verdicts both stable and efficient.

To demonstrate the framework, the researchers applied it to the embankment of the Helong Reservoir in Guangzhou City, in China’s Guangdong Province. The model’s output is expressed as a membership degree, a measure of how strongly the embankment’s overall condition belongs to each safety grade. The result showed that the membership degree of the Helong Reservoir to the normal category reached 62.9 percent, a clear indication that the structure was operating in a satisfactory state. Crucially, this verdict was consistent with the evaluation results of other established models and with the actual current situation of the reservoir, confirming the validity and reasonableness of the new method.

The agreement between the game theory-cloud model and alternative evaluation approaches is more than a technical footnote. Dam safety evaluation models are needed to accurately identify the safety status of dams precisely because failure can have a significant impact on the lives and health of people downstream. Statistical analyses of dam failures in China, cited in the study’s bibliography, underline how monitoring data, inspection records, and formal review indicators each tell part of the story, and how a comprehensive model that weighs all three can catch deterioration before it becomes an emergency. The new framework provides a reference for the comprehensive evaluation of embankment safety that other agencies can adapt to their own structures.

Behind the mathematics lies a broader movement in dam engineering toward data-driven health diagnosis. Recent studies have explored multi-source monitoring information fusion based on Wasserstein distance, anomaly identification for tailings dams, deep learning methods for dam behavior after extreme loads, and even digital twins that mirror a dam’s condition in real time. The cloud model has already been applied to dam safety monitoring, water cycle health assessment, and risk evaluation of dam-break environmental impacts. What distinguishes the new contribution is the systematic way it merges subjective and objective weighting through improved game theory while simultaneously taming the randomness inherent in cloud-based reasoning, two improvements that address the main sources of error in prior cloud-model evaluations.

The practical implications reach far beyond one reservoir in Guangdong. Embankments and earthen dams number in the tens of thousands worldwide, many of them aging structures whose instrument networks are sparse and whose inspection regimes depend on human judgment. A model that can quantitatively reconcile expert opinion with measured data, and that can express its conclusion as a graded membership rather than a binary pass or fail, gives operators a richer picture of margin and risk. The 62.9 percent membership in the normal category for Helong Reservoir illustrates the granularity available: instead of a single label, managers see how strongly the evidence supports each safety state, which can guide the intensity of future monitoring and inspection.

The research was supported by the National Natural Science Foundation of China, the Science and Technology Innovation Program from Water Resources of Guangdong Province, and the Guangdong Provincial Water Conservancy and Hydropower Technology Center. As climate change intensifies floods and raises the stakes for water infrastructure, tools that turn uncertain evidence into clear safety verdicts are likely to attract growing attention from engineers and regulators alike. The improved game theory-cloud model joins a rapidly expanding toolkit for keeping the world’s aging embankments under quantitative watch, and its validation at Helong Reservoir suggests the approach is ready to be tested on structures whose safety is far less certain.

Subject of Research: Comprehensive Evaluation of Embankment Safety Based on Improved Game Theory-Cloud Model

Article Title: Comprehensive Evaluation of Embankment Safety Based on Improved Game Theory-Cloud Model

Article References: Zhang, J., Wu, W., Hou, G., Huang, J., & Jia, B. (2026). Comprehensive Evaluation of Embankment Safety Based on Improved Game Theory-Cloud Model. Water Resources Management, 40(14), Article 540. https://doi.org/10.1007/s11269-026-04907-6

Image Credits: AI Generated

DOI: 10.1007/s11269-026-04907-6

Keywords: Comprehensive, Evaluation, Embankment, Safety, Based, Improved, Game, Theory-Cloud, Model, scientific research