(610f) Proactive Quasi-Decentralized Networked Model Predictive Control

The fundamental and practical challenges associated with control of networked process systems have been the focus of significant research work in process control over the past two decades, and have motivated many research studies in this area. Examples include results on traditional centralized and decentralized control, distributed model predictive control (e.g., [1]-[2]), passivity-based control (e.g., [3]), agent-based systems (e.g., [4]), and multi-time scale control (e.g., [5]). In addition to these results, research efforts within process control have also begun to address the new challenges that arise from the use of networked control systems in process operations (e.g., resource constraints, real-time scheduling constraints, etc.). Efforts to address some of these issues were initiated in [6]-[8] and focused on the development of resource-aware networked plant-wide control, estimation and scheduling strategies with explicitly characterized closed-loop stability properties.

These results were subsequently extended in [9] to incorporate performance optimization considerations through the use of model predictive control techniques. Specifically, an adaptive quasi-decentralized MPC framework, where each subsystem is controlled by a Lyapunov-based MPC controller, was developed, and a forecast-triggered communication strategy was used to minimize the cross communication of measurements between the distributed subsystems. This communication strategy exploits the stability properties of each subsystem and generates forecasts of the future behavior of each subsystem locally. When the forecast indicates a potential significant loss of control performance for a particular subsystem, its corresponding control system sends out requests to all the other subsystems for measurements which are used to update the model states within the local MPC controller. In this manner, a balance can be maintained between the performance of each subsystem and the utilization of network resources. Given that this adaptive (i.e., request-based) strategy requires that the local control system broadcast requests to all the other subsystems simultaneously and then await the arrival of measurements from the entire plant within a possibly short period of time, this could potentially lead to unnecessary network utilization which might lead to communication delays and data losses.

To address this problem, we present in this work a quasi-decentralized MPC framework that combines proactive and adaptive communication strategies. A set of local MPC controllers are initially designed based on sampled-data distributed bounded Lyapunov-based controllers, and their closed-loop stability and performance properties are characterized in terms of local (subsystem-specific) and global (plant-wide) thresholds. A hybrid communication strategy with complementary proactive and adaptive components is then devised to meet the control and communication objectives at the local and plant-wide levels. In both components, each subsystem forecasts the future evolution of the local state within the next sampling interval (relative to the corresponding stability/performance thresholds) and then decides whether to establish or suspend communication with the rest of the plant. In the proactive strategy, the forecasted breach of the specified threshold triggers each subsystem to broadcast its measurements to update the model states within the other subsystems to maintain the overall plant stability; while in the adaptive strategy the subsystem responds to the local threshold breach by requesting measurements from the other subsystems to update the states of the models embedded within the subsystem itself to maintain the local subsystem stability.

The proactive broadcast-based communication strategy is implemented when the process state of the entire plant is far from the nominal equilibrium point and its corresponding control objective is to stabilize the overall system and drive its state to some neighborhood of the equilibrium point. As the process state converges to this neighborhood, the contribution of the local behavior to the overall plant stability becomes difficult to discern, and thus the proactive strategy becomes infeasible. Under such circumstances, all control systems switch to the adaptive request-based communication strategy where the control objective becomes the practical stabilization of every subsystem. A simulated chemical process network example is used to illustrate the implementation of the proposed approach.

References:

[1] B. T. Stewart, S. Wright and J. B. Rawlings, "Cooperative distributed model predictive control for nonlinear systems," J. Process Control, 21, 698–704, 2011.

[2] P. D. Christofides, J. Liu and D. M. de la Pena. Networked and Distributed Predictive Control: Methods and Nonlinear Process Network Applications. Springer-Verlag, London, 2011.

[3] K. R. Jillson and B. E. Ydstie, "Process networks with decentralized inventory and flow control," J. Process Control, 17, 399-413, 2007.

[4] M. D. Tetiker, A. Artel, F. Teymour and A. Cinar, A., "Control of grade transitions in distributed chemical reactor networks: An agent-based approach," Comp. Chem. Eng., 32,1984-1994, 2008.

[5] S. Jogwar, M. Baldea and P. Daoutidis, "Dynamics and control of process networks with large energy recycle, " Ind. Eng. Chem. Res., 48, 6087-6097, 2009.

[6] Y. Sun and N. H. El-Farra, "Quasi-decentralized model-based networked control of process systems,'' Comp. Chem. Eng., 32, 2016-2029, 2008.

[7] Y. Sun and N. H. El-Farra, "A quasi-decentralized approach for networked state estimation and control of process systems," Ind. Eng. Chem. Res., 49, 7957-7971, 2010.

[8] Y. Sun and N. H. El-Farra, "Resource-aware quasi-decentralized control of networked process systems over wireless sensor networks," Chem. Eng. Sci., 69, 93-106, 2012.

[9] Y. Hu and N. H. El-Farra, "Adaptive quasi-decentralized model predictive control of networked process systems," Distributed MPC Made Easy, R. Negenborn and P. Maestre (Eds.), Chapter 12, 209-223, Springer-Verlag, Berlin, 2014.