(359a) Optimization-Based Predictive Control of Networked Process Systems with Discrete and Delayed Sensor-Controller Communication

A key problem in networked control systems is the design of resource-aware control methods that can systematically balance the inherent tradeoff between the desired closed-loop stability and performance requirements on the one hand and the constraints on the sensing, computation and communication resources of the networked devices on the other. It is well known that maximizing closed-loop system performance often requires frequent sensor-controller communication and increased levels of network resource utilization, whereas reducing communication costs favors keeping the sensing and communication levels to a minimum which can compromise the control system performance.

An approach that seeks to enforce closed-loop stability with reduced sensor-controller communication is model-based networked control [1]. The basic idea of this approach is to embed a predictive model of the plant within the control system to generate the control action when sensor-controller communication over the network is suspended and state measurements are unavailable to the controller, and to update the model state when communication is restored. Through a closed-loop stability analysis, an explicit characterization of the minimum allowable model update frequency needed to achieve closed-loop stability can be obtained. This approach has subsequently been extended and applied to address a wide range of problems, including control of large-scale process networks [2], distributed control of diffusion-reaction processes with spatially-distributed sensors and actuators [3], and fault-tolerant stabilization of networked process systems [4]. While these approaches help reduce network resource utilization without compromising closed-loop stability, a close examination of existing model-based networked control strategies reveals a general lack of explicit control performance optimization in the controller design framework. In the context of balancing the intrinsic tradeoff between control performance and network resource utilization, there has been an apparent increased emphasis on reducing network utilization, often at the expense of the achievable control system performance.

An effort to address this issue was made in [5], [6] where performance optimization considerations and control constraints were explicitly incorporated in the networked controller design methodology through the use of model predictive control formulations. The focus of these studies, however, has been on optimizing the closed-loop performance, with respect to the response speed and the control effort, without explicit characterization or consideration of the sensor-controller communication cost. The problem of reducing sensor-controller communication costs was addressed indirectly through the use of a separate event-triggered control approach.

In this work, we present a methodological framework for the design and implementation of optimization-based predictive control systems for networked processes with discrete and delayed sensor-controller communication over a resource-constrained network. The framework aims to optimize the control system performance, while simultaneously minimizing sensor-controller communication costs and overall network utilization. Initially, an auxiliary model-based networked controller with an embedded propagation unit that compensates for the communication delay is designed. The closed-loop stability region is explicitly characterized in terms of the controller gain, the communication delay and the maximum allowable model update period that ensures closed-loop stability under periodic sensor-controller communication. Based on this characterization, a finite-horizon optimization problem in which the cost function imposes explicit penalties on the model state, the control action and the model update frequency is formulated subject to the delay-dependent stability constraints. The optimization problem is solved in a receding horizon fashion to determine the optimal controller gain and optimal model update period needed at each update time. The receding horizon strategy results in time-varying controller gain and communication scheduling policies that allow the process to adapt more effectively to environmental uncertainties and perturbations. The implementation and efficacy of the developed optimization-based networked control approach are illustrated using a chemical process example.

References:

[1] L. A. Montestruque and P. J. Antsaklis, “On the model-based control of networked systems,” Automatica, vol. 39, pp. 1837–1843, 2003.

[2] D. Xue and N. H. El-Farra, “Supervisory logic for control of networked process systems with event-based communication,” in Proceedings of American Control Conference, Boston, MA, 2016, pp. 7135–7140.

[3] Z. Yao and N. H. El-Farra, “Model-based networked control of spatially distributed systems with measurement delays,” in Proceedings of American Control Conference, Montreal, Canada, 2012, pp. 2990–2995.

[4] Z. Yao and N. H. El-Farra, “Data-driven actuator fault identification and accommodation in networked control of spatially-distributed systems,” in Proceedings of American Control Conference, Portland, OR, 2014, pp. 1021–1026.

[5] Z. Yao and N. H. El-Farra, “Resource-aware model predictive control of spatially distributed processes using event-triggered communication,” in Proceedings of 52nd IEEE Conference on Decision and Control, Florence, Italy, 2013, pp. 3726–3731.

[6] Y. Hu and N. H. El-Farra, “Adaptive quasi-decentralized MPC of networked process systems,” in Distributed Model Predictive Control Made Easy, vol. 69, Maestre, J. M. and Rudy R. Negenborn (Eds.), Springer: Berlin, 2014, pp. 209–223.