Energy-based Adaptive Compression in Water Network Control Systems Sokratis Kartakis * , Marija Milojevic Jevric * , George Tzagkarakis †‡ , and Julie A. McCann * * Department of Computing, Imperial College London, UK E-Mails:{s.kartakis13, m.milojevic-jevric, j.mccann}@imperial.ac.uk † EONOS Investment Technologies, Paris, France;E-Mail: gtzag@eonos.com ‡ Foundation for Research and Technology-Hellas, Institute of Computer Science/SPL, Greece Abstract—Contemporary water distribution networks ex- ploit Internet of Things (IoT) technologies to monitor and con- trol the behavior of water network assets. Smart meters/sensor and actuator nodes have been used to transfer information from the water network to data centers for further analysis. Due to the underground position of water assets, many water companies tend to deploy battery driven nodes which last beyond the 10-year mark. This prohibits the use of high-sample rate sensing therefore limiting the knowledge we can obtain from the recorder data. To alleviate this problem, efficient data compression enables high-rate sampling, whilst reducing significantly the required storage and bandwidth resources without sacrificing the meaningful information content. This paper introduces a novel algorithm which combines the ac- curacy of standard lossless compression with the efficiency of a compressive sensing framework. Our method balances the tradeoffs of each technique and optimally selects the best compression mode by minimizing reconstruction errors, given the sensor node battery state. To evaluate our algorithm, real high-sample rate water pressure data of over 170 days and 25 sensor nodes of our real world large scale testbed was used. The experimental results reveal that our algorithm can reduce communication around 66% and extend battery life by 46% compared to traditional periodic communication techniques. Keywords-IoT, Cyber-Physical Systems, Wireless Sensor and Actuator Networks, Smart Water Network, Compressive Sens- ing I. I NTRODUCTION Optimal water distribution and energy waste reduction are currently hot topics. Water demands are not being met in many regions around the globe; both developed and underdeveloped; where climate change and economic water scarcity are two issues that have the largest impact. Notwith- standing the 7.5bn investment in UK water distribution networks, 3.3bn liters of water were lost per day in 2010 [1]. In order to decrease maintenance costs and water waste, recent years, water utility companies increasingly transform their old water distribution networks to smart by exploiting Information Communication Technologies (ICT). Current systems exploit energy hungry over ground deployments to monitor and transmit water network states (i.e. water flow and pressure) to a server periodically -typically via the mobile phone networks- in order to detect anomalous behaviors such as water leakage and bursts [2], [3], [4], [5], [6]. However, more than 97% of water network assets are located away from power resources and in geographically remote unpopulated areas; which make current approaches unsuitable for next generation smart water networks. Battery-driven wireless sensor networks are a strong so- lution for these large-scale smart water systems. The main challenge of this approach is that sensor nodes require a lot of energy to transmit high precision data, which is required for accurate anomaly (i.e. burst and leakages) detection algorithms. To address this problem, we have proposed two solutions to reduce data volume: (a) lossless compression [7] and (b) lossy compression by using the powerful framework of compressive sensing [8], which both of them minimize the communication by covering high information level needs to the server-side. Each compression technique has tradeoffs, which has been analyzed in [8]. Specifically, in lossless compression, the initial stream can be reconstructed completely without losing information, while compressive sensing introduces a recon- struction error. On the other hand, compressive sensing can significantly reduce the amount of data, and consequently the communication. Lossless compression places an upper bound on the compression performance. This paper balances these tradeoffs and presents an opti- mal algorithm which selects the best compression mode in a dynamic and distributed fashion by minimizing the recon- struction error, given the current sensor node battery state. Based on evaluation results with real data, this algorithm can reduce the communication by around 66% and extend the battery life by 46% compared to the traditional periodic communication approaches. The rest of the paper is organized as follows: Section II overviews the the system architecture, along with pre- liminary concepts. Section III formulates the optimization problem, while Section IV analyses our proposed algorithm for the optimal selection of the compression mode. In Section V, the performance of our method is evaluated on a real data set, while Section VI summarizes the main results. II. SYSTEM OVERVIEW AND PRELIMINARIES Water utility companies deploy sensor nodes in contempo- rary smart water networks, such as [2], to monitor network states. These sensor nodes transmit high sample rate flow and pressure data periodically in 30-second and 5-minute intervals respectively through 3G or WiFi to a server which