Scholarlink

<p>Efficient management of water resources is crucial in the face of urbanization, climate change, and increasing population densities. This paper introduces a privacy-preserving framework for short-term water consumption forecasting in smart cities, leveraging Federated Learning (FL) and Long Short-Term Memory (LSTM) networks. The proposed approach uses FL to ensure that the data remain local to each consumer, preserving privacy while training predictive models. We investigate two types of FL models: a global model trained across multiple households, and a local model personalized to individual consumption behaviors. Our results demonstrate that the local model significantly improves prediction accuracy compared to the global model by capturing unique household-level consumption patterns. Moreover, we introduce an incentive mechanism designed to align consumer behavior with predicted consumption, rewarding users whose water usage closely matches the forecast. This incentive mechanism promotes efficient and consistent water use, ultimately reducing the discrepancies between predicted and actual consumption. Extensive simulations show that the personalized model yields more accurate predictions and enhances user engagement through tailored incentives. This research contributes to bridging advanced machine learning methodologies with practical implementations in urban resource management, emphasizing privacy, scalability, and user-centric incentives as key components of sustainable water consumption in modern smart cities.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Federated learning with user incentives for enhanced resource consumption forecasting in urban IoT systems

  • 期刊论文

  • 发表日期：

    2026年1月4日

查看全文
  • Mohammed El Hanjri,
  • Amine Abouaomar,
  • Anas Abouaomar,
  • Abdellatif Kobbane

摘要

Efficient management of water resources is crucial in the face of urbanization, climate change, and increasing population densities. This paper introduces a privacy-preserving framework for short-term water consumption forecasting in smart cities, leveraging Federated Learning (FL) and Long Short-Term Memory (LSTM) networks. The proposed approach uses FL to ensure that the data remain local to each consumer, preserving privacy while training predictive models. We investigate two types of FL models: a global model trained across multiple households, and a local model personalized to individual consumption behaviors. Our results demonstrate that the local model significantly improves prediction accuracy compared to the global model by capturing unique household-level consumption patterns. Moreover, we introduce an incentive mechanism designed to align consumer behavior with predicted consumption, rewarding users whose water usage closely matches the forecast. This incentive mechanism promotes efficient and consistent water use, ultimately reducing the discrepancies between predicted and actual consumption. Extensive simulations show that the personalized model yields more accurate predictions and enhances user engagement through tailored incentives. This research contributes to bridging advanced machine learning methodologies with practical implementations in urban resource management, emphasizing privacy, scalability, and user-centric incentives as key components of sustainable water consumption in modern smart cities.