RESEARCH

InterCONnected Critical Infrastructure Systems

Engineering (CONCISE) Laboratory

Ongoing Research:

• Dynamic State Estimation of Smart Water Distribution Systems with Low Observability

• Data-driven control of cyber-physical infrastructure systems

• Cybersecurity of Critical Interdependent Systems

• Model Predictive Control of Critical Interdependent Systems

• Resilience Assessment and Water Demand Forecasting of Water Distribution Systems 

Data-driven control of cyber-physical infrastructure systems

Considering the complex nature of WDSs, providing optimal control of WDSs is crucial for delivering high-quality and reliable service to consumers. This goal can be accomplished by utilizing a mathematical model to control WDS. However, deriving physics-based dynamic models is arduous, particularly for large-scale. Therefore, the CONCISE lab continues to investigate different data-driven dynamic model identification techniques that can be applied for control purposes, thereby eliminating the reliance on the WDSs' physical models.  One of our published studies integrated data-driven dynamic identification that leverages sparse-regression theory (SR-based ID) with three distinct MPC techniques: 1) linear time-invariant, 2) linear time-varying via successive linearization, 3) and nonlinear MPCs. This work is published in the Journal of Water Process Engineering. In another study, data-driven system identification was conducted using kernel-based interpolation, assuming general WDS dynamics without prior knowledge of the true basis functions. Leveraging WDSs' automation, such as water level and flow sensors, this method constructs a regularized interpolated kernel-based model based on input-output pairs. The results demonstrate high accuracy with errors varying from 0.02% to 3%. This work was presented at the EWRI Congress 2024 and received the Third Best Paper Award in the Graduate Paper Competition. Both studies are validated on four interconnected water tanks, representing simplified WDSs yet equipped with WDSs’ nonlinearities. 

Cyberattack Localization and Detection in Smart Water Distributions Systems | Optimization & Machine Learning

Due to the increasing sophistication of cyberattacks on critical infrastructures, the CONCISE lab has been investigating the cyber-vulnerability of water systems using chance-constrained optimization algorithms. In one study, a small water network was simulated to analyze the consequences of stealthy cyberattacks on vulnerable nodes and junctions.  Another study examined the combined impact of cyberattacks, unknown disturbances, and other uncertainties on the water system using a nonlinear optimization framework. To develop an advanced cyberattack detection mechanism, we introduced an attack detection framework employing deep neural networks, specifically Autoencoders. This work was presented at the EWRI Congress 2024 and received the Second Best Paper Award in the Graduate Paper Competition. The goal of these efforts is to enhance the cybersecurity of smart water distribution systems in the face of emerging cyber threats.

Model Predictive Control of Critical Interdependent Systems

Water distribution systems (WDSs) are inherently complex, uncertain, and nonlinear. Ineffective management can lead to unmet water demand, system failure, and delivery disruptions. This motivates the CONCISE Lab to determine the optimal control strategy to ensure reliable WDS operation. Model predictive control (MPC) emerges as the preferred control strategy due to its predictive capability and efficacy in handling multi-objectives and multivariable constraints tailored to WDS characteristics. However, MPC is limited by its high computation costs to yield optimal control solutions. In our recent study, we proposed a fast computation algorithm that utilizes an interpolated move-blocking strategy to reduce the computational burden that MPC entails. The objective is to confine the degrees of freedom (DoF) of the control inputs in a manner that prevents continuous variation of the control signal. Fast MPC performed 1) a successful demand-driven and cost-effective WDS operation, providing optimal pump scheduling, and 2) achieved fast computation, reducing 80\% computation time with consistent optimality convergence. 

Resiliency Assessment and Water Demand Forecasting of Water Systems | Optimization & Machine Learning

To better inform the design and development of improvement alternatives for water network optimization and management processes, the Concise Lab continues to investigate how the application of already existing machine-learning algorithms and the design of novel algorithms and techniques can improve water demand forecasting capabilities for water networks. One of our published studies developed a novel hybrid demand forecasting model for a real-world water demand dataset to account for system uncertainties using a probabilistic prediction framework. This work is published in the Journal of Water Process Engineering.

Moreover, specific stressors or catastrophic events could hinder networks from fulfilling their demand requirements and lead to their eventual failure. Thus, our research efforts considerably cover the assessment and quantification of water networks' resilience against such stressors. In one study, we developed a composite resilience metric based on the resilience curve paradigm to describe both the disruption and recovery phases of water networks. This work was presented at the EWRI Congress 2024 and continues to undergo further investigation. 

Stealthy cyberattack injection and detection using nonlinear programming.

Previous efforts were focused on two important aspects of cybersecurity in water distribution systems: (1) Developing models and test cases to study various types of cyberattacks in water distribution networks, and (2) detection of cyberattacks that are deliberately designed to fail components of water distribution network or cause cascading failures. In the first aspect, we focus on designing and analyzing stealthy false data injection (FDI) cyberattack models that can bypass bad data detection algorithms to help authorities develop more practical, precise and timely countermeasures. So far, we have focused on modeling stealthy cyberattack in water distribution systems that can cause 1) nodal head increase and cascading failures, and 2) tanks' overflown or withdrawn. The modeling uses a bi-level nonlinear programming formulation that accounts for false data injections that could successfully bypass state-estimation algorithm without being detected by the operator. 

In a separate effort to detect cyberattacks in water distribution networks, we developed optimization models using MINLP formulation that computes nodal demands and pressure heads in WDS to be compared with those obtained from the state-estimation algorithm, identifying potential FDI attacks. In another research, we work on developing an integrated state-estimation framework to detect potential FDIs in a combined water-energy system (shown in the figure below). 

This research was supported by funding from Center for Security Research and Education to use machine learning algorithms to detect cyberattacks in water and energy networks.