[Paper Review] Federated Learning for Water Consumption Forecasting in Smart Cities

Mohammed El Hanjri, Hibatallah Kabbaj+2 more|arXiv (Cornell University)|Jan 30, 2023

Smart Parking Systems Research4 citations

TL;DR

This paper proposes a federated learning (FL)-based approach for short-term water consumption forecasting in smart cities, training Long Short-Term Memory (LSTM) models locally on edge devices (e.g., smart meters) to preserve user privacy and reduce network load. The method achieves higher prediction accuracy than centralized models while minimizing data transmission and protecting sensitive consumption patterns through decentralized training and model aggregation.

ABSTRACT

Water consumption remains a major concern among the world's future challenges. For applications like load monitoring and demand response, deep learning models are trained using enormous volumes of consumption data in smart cities. On the one hand, the information used is private. For instance, the precise information gathered by a smart meter that is a part of the system's IoT architecture at a consumer's residence may give details about the appliances and, consequently, the consumer's behavior at home. On the other hand, enormous data volumes with sufficient variation are needed for the deep learning models to be trained properly. This paper introduces a novel model for water consumption prediction in smart cities while preserving privacy regarding monthly consumption. The proposed approach leverages federated learning (FL) as a machine learning paradigm designed to train a machine learning model in a distributed manner while avoiding sharing the users data with a central training facility. In addition, this approach is promising to reduce the overhead utilization through decreasing the frequency of data transmission between the users and the central entity. Extensive simulation illustrate that the proposed approach shows an enhancement in predicting water consumption for different households.

Motivation & Objective

• To address privacy concerns in smart city water consumption forecasting by avoiding central collection of individual household data.
• To improve prediction accuracy for individual households using localized, personalized deep learning models.
• To reduce network overhead by minimizing data transmission through edge-based federated learning.
• To evaluate the performance of FL in comparison to centralized training in terms of accuracy and communication efficiency.
• To demonstrate the feasibility of deploying FL with edge devices for real-time, privacy-preserving water demand forecasting.

Proposed method

• The system uses a federated learning framework where a global LSTM model is trained across multiple edge devices (e.g., smart meters) without sharing raw consumption data.
• Each client (household) trains the global model locally on its own monthly water consumption data using a personalized local update.
• Model weights are transmitted to a central server, where they are averaged to update the global model, ensuring privacy through data decentralization.
• The process iterates over multiple rounds until convergence, with the global model improving through diverse, localized data.
• The approach leverages edge computing infrastructure (e.g., home grid Edge technology) to enable on-device training and reduce latency.
• Network load is quantified using equations for centralized ($Q_C$) and federated ($Q_F$) training, with a reward metric $R = 1 - Q_F/Q_C$ to measure communication efficiency gains.

Figure 1: Network architecture and components

Experimental results

Research questions

• RQ1 Can federated learning improve short-term water consumption forecasting accuracy compared to centralized deep learning models?
• RQ2 To what extent does local model personalization enhance prediction performance for individual households?
• RQ3 How does federated learning reduce network load compared to centralized training in smart water systems?
• RQ4 What is the impact of decentralized training on data privacy in residential water consumption forecasting?
• RQ5 Can edge-based FL be practically deployed for real-time, scalable water demand prediction in smart cities?

Key findings

• The proposed federated learning approach achieved higher prediction accuracy than the centralized global model, particularly for personalized forecasts on individual households.
• Local retraining on client-specific data significantly improved forecasting performance, as demonstrated by closer alignment between predicted and actual consumption in BLD 04 (2018–2020).
• The network load reduction reward $R$ was positive, indicating that federated learning reduced data transmission by a measurable margin compared to centralized training.
• The model updates were transient and never stored on the server, minimizing privacy risks and preventing reverse engineering of user identities.
• Even for new users with limited data, the global model provided a viable baseline prediction, demonstrating robustness and scalability.
• The integration of edge computing enabled on-device training, making the system feasible for deployment on real-world IoT devices in smart water infrastructure.

Figure 2: The distribution of water consumption data in building 04 (BLD 04). (HCF; Hundred-Cubic Foot)