Formal controller synthesis for wastewater systems with signal temporal logic constraints: The Barcelona case study
Introduction
The infrastructure for water and wastewater management is being continuously upgraded due to the constant increase in demand for water and wastewater services as a result of population growth. In order to support this upgrade, the water industry has been investigating the potential benefits of using more advanced automatic control strategies. The design and automatic control of sewer networks pose new challenges to the control community. The newly designed methodologies should be able to handle the effect of uncertainties in the amount of precipitation, the physical and operational constraints of the network, and the effects of delays and nonlinearities in the dynamics of the system. These challenges require improving performance of the traditional control strategies such as on-off and PID controllers, which are not capable of handling such issues. Model predictive control (MPC) seems to be a suitable methodology to control sewer networks as it can deal with these particular challenges associated with such systems. MPC is an online control technique that uses a mathematical model of the considered system to compute the control inputs by minimizing a cost function [1], [2], [3], [4]. Moreover, it is capable of incorporating either linear or nonlinear dynamics of the system as well as handling constraints on inputs, states and outputs. Hence, the MPC methodology is quite suitable for the global control of urban sewage systems within a hierarchical control structure [5], [6].
The system under investigation in this paper is part of the Barcelona wastewater system, which is subject to sudden weather-change events within the Mediterranean climate. We consider the Barcelona test catchment (BTC) that covers a surface area of 22.6 km2 and represents all the typical elements of the whole network. The application of deterministic MPC to Barcelona wastewater system has been investigated in [7] for a portion of this system and its benefits have been examined toward the potential percentage reductions in both flooding and pollution in Barcelona sewage network. In this paper, we build on the work of [7] by including uncertainty in the amount of precipitation as a bounded disturbance and by formulating a robust MPC optimization problem [8], [9], [10], [11], [12] to synthesize control inputs.
In order to specify the desired behavior of a system with continuous dynamics, signal temporal logic (STL) is one of the most useful languages. In comparison with other temporal logic formalisms, STL has the advantage of naturally admitting a quantitative semantics. As such, in addition to the binary answer to the satisfaction question of the specifications, it provides a real number that indicates the extent to which the specification is either satisfied or violated. This quantitative semantics associated to the STL specification is referred to as the robustness function. Incorporating such temporal specification in the optimization problem formulation enforces the closed-loop system to satisfy the desired temporal behavior, as it is confirmed by the simulation results of this paper.
Considering the nonlinear (or hybrid) nature of the network model, we show that the proposed robust MPC optimization problem can be formulated as mixed integer linear or quadratic programming (MILP or MIQP) problems as follows. First, the nonlinear dynamics of the wastewater network are transformed into a mixed logical dynamical (MLD) model. Then, the nonlinear expressions in the objective function and the STL constraints are transformed into mixed integer linear terms and constraints, respectively. Finally, we employ either dual reformulation or Monte Carlo method in the inner optimization problem, i.e., the maximization problem, to get either an MIQP problem or an MILP problem. In the case of MIQP, we obtain a non-convex optimization problem, which we solve iteratively by linear approximation of the quadratic objective function. In the simulation results, we compare the performance of the dual reformulation with the Monte Carlo approach and we show the effect of STL specifications on system behavior.
STL has been used for controller synthesis in a variety of domains for uncertain systems using receding horizon control techniques [13], [14], [15]. Transforming STL constraints into mixed integer linear constraints has been used in [16]. Several works related to this wastewater system consider different models and cope with the design of alternative MPC approaches, e.g., [17], [18] and references therein. Recent works have proposed different approaches for handling uncertainties in process control. The work reported in [19] proposes a two-level method to first estimate the worst-case disturbance profile using an uncertain finite impulse response (FIR) model. This profile is then employed to simulate the closed-loop nonlinear dynamic process model for obtaining the worst-case output variability and checking the feasibility of constraints. Likewise, the work reported in [20] proposes an MPC strategy that relies on nonlinear optimizations. This approach incorporates integer variables towards performing a modeling selection within the control structure. In our previous work [21], we studied a small part of BTC with only 3 tanks. In the current paper, we consider the full model of BTC, as presented in [7], to show that our method is both scalable and efficient for formally synthesizing control inputs for the system.
Our work is distinct from the previous works on wastewater systems in (a) considering uncertainty in the amount of precipitation both in the model and in the controller design; (b) employing STL to encode desired properties of the closed-loop trajectories; (c) proposing an approximate solution for the formulated optimization problem that is scalable and can be applied to the large dimensional model of the BTC.
The remainder of this paper is organized as follows. In Section 2, the considered model of the BTC is described. In Section 3, the robust MPC formulation is presented together with the constraints induced by both the model of the system and the STL specifications. In Section 4, we discuss the MLD model of the system and propose solution approaches to solve the mixed integer robust MPC optimization problem. In Section 5, the proposed control approach is applied to the BTC and the main results are proposed and discussed. Finally, Section 6 draws the main conclusions of the paper and the possible lines of future research. In order to keep the discussion of the paper focused, we summarize STL semantics and the notion of robustness in the appendices.
Section snippets
Barcelona test catchment model
We consider a portion of the sewer network of Barcelona that is representative, as it exhibits the main phenomena and the most common characteristics found in the entire network. The network consists of nine tanks, four control inputs corresponding to the manipulated flows, and eleven measured disturbances corresponding to the measurements of rain precipitation. Two wastewater treatment plants (WWTP) are used to treat the sewage before it is released to the receiving environment. Fig. 1 shows
Robust model predictive control
In the BTC, the goal is to control the inflow and outflow in (both virtual and real) tanks in order to avoid flooding and contaminating Mediterranean sea. The uncertainty in the wastewater system is in the amount of precipitation that we consider to be a bounded quantity. The control objectives are to minimize both flooding on streets (overflows and q12Sc) and the pollution entering the sea (overflows q9c and and flows q7, q8, q10, and q11 in Fig. 1) as well as
Solving the robust MPC problem
In order to solve the optimization problem (8), we transform the hybrid system model into its equivalent MLD form. An MLD model is a linear system model with both continuous and binary variables while having affine constraints on these variables. The MLD formalism allows the transformation of logical statements involving continuous variables into mixed integer linear inequalities. We employ the following equivalences [23] to transform the nonlinear dynamics of the system and nonlinear terms in
Simulation results
We apply our proposed synthesis techniques to the model of the BTC presented in Fig. 1. The system's matrices for the MLD representation of the network (10) are given in Appendix B. We have chosen the sampling time Δt = 300s, according to the evolution of the real system. The chosen sampling time is related to the time of concentration2 determined for the BTC by
Conclusion
We proposed an effective control approach to manage the flow in a representative fragment of Barcelona wastewater network. We formulated the nonlinear dynamics of the flow network including the possible overflows in the considered catchment. The goal of the control scheme is to minimize the overflow in the main pipes and to maximize the amount of water treatment. We have employed model predictive control (MPC) to optimally manage the flow into the network. We have used signal temporal logic
Acknowledgements
The work of C. Ocampo-Martinez is partially supported by the project DEOCS (Ref. DPI2016-76493-C3-3-R) from the Spanish MINECO/FEDER.
References (33)
- et al.
Real time control of urban wastewater systems: where do we stand today
J. Hydrol.
(2004) - et al.
Robust constrained model predictive control using linear matrix inequalities
Automatica
(1996) - et al.
Robust model predictive control of constrained linear systems with bounded disturbances
Automatica
(2005) - et al.
An MPC-based control structure selection approach for simultaneous process and control design
Comput. Chem. Eng.
(2014) - et al.
On hybrid systems and closed-loop MPC systems
IEEE Trans. Autom. Control
(2002) - et al.
Stability of hybrid model predictive control
IEEE Trans. Autom. Control
(2006) Predictive Control with Constraints
(2002)- et al.
Model Predictive Control: Theory and Design
(2009) - et al.
Optimal Real-Time Control of Sewer Networks
(2005) Model Predictive Control of Wastewater Systems
(2010)
On min-max model-based predictive control
Advances in Model-Based Predictive Control
Robust model predictive control of stable linear systems
Int. J. Control
Robust model predictive control
Proceedings of American Control Conference (ACC), vol. 2
Shrinking horizon model predictive control with chance-constrained signal temporal logic specifications.
Robust model predictive control for signal temporal logic synthesis.
Reactive synthesis from signal temporal logic specifications.
Cited by (10)
Adaptive critic design with weight allocation for intelligent learning control of wastewater treatment plants
2024, Engineering Applications of Artificial IntelligenceModeling and control of wastewater treatment process with time delay based on event-triggered recursive least squares
2023, Engineering Applications of Artificial IntelligenceData-driven verification under signal temporal logic constraints
2020, IFAC-PapersOnLineEvent-Driven Model Predictive Control With Deep Learning for Wastewater Treatment Process
2023, IEEE Transactions on Industrial InformaticsFormally Verified Self-adaptation of an Incubator Digital Twin
2022, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
- 1
The main part of the research is performed while Sadegh Soudjani was with Max Planck Institute, Germany.