Design and implementation of predictive control for networked multi-process systems

Abstract

This thesis is concerned with the design and application of the prediction method in the NMAS (networked multi-agent system) external consensus problem. The prediction method has been popular in networked single agent systems due to its capability of actively compensating for network-related constraints. This characteristic has motivated researchers to apply the prediction method to closed-loop multi-process controls over network systems. This thesis conducts an in-depth analysis of the suitability of the prediction method for the control of NMAS. In the external consensus problem, NMAS agents must achieve a common output (e.g. water level) that corresponds to the designed consensus protocol. The output is determined by the external reference input, which is provided to only one agent in the NMAS. This agreement is achieved through data exchanges between agents over network communications. In the presence of a network, the existence of network delay and data loss is inevitable. The main challenge in this thesis is thus to design an external consensus protocol with an efficient capability for network constraints compensation. The main contribution of this thesis is the enhancement of the prediction algorithm’s capability in NMAS applications. The external consensus protocol is presented for heterogeneous NMAS with four types of network constraints by utilising the developed prediction algorithm. The considered network constraints are constant network delay, asymmetric constant network delay, bounded random network delay, and large consecutive data losses. In the first case, this thesis presents the designed algorithm, which is able to compensate for uniform constant network delay in linear heterogeneous NMAS. The result is accompanied by stability criteria of the whole NMAS, an optimal coupling gains selection analysis, and empirical data from the experimental results. ‘Uniform network delay’ in this context refers to a situation in which the agent experiences a delay in accessing its own information, which is identical to the delay in data transfer from its neighbouring agent(s) in the network In the second case, this thesis presents an extension of the designed algorithm in the previous chapter, with the enhanced capability of compensating for asymmetric constant network delay in the NMAS. In contrast with the first case—which required the same prediction length as each neighbouring agent, subject to the same values of constant network delay—this case imposed varied constant network delays between agents, which required multi-prediction lengths for each agent. Thus, to simplify the computation, we selected a single prediction length for all agents and determined the possible maximum value of the constant network delay that existed in the NMAS. We tested the designed control algorithm on three heterogeneous pilotscale test rig setups. In the third case, we present a further enhancement of the designed control algorithm, which includes the capability of compensating for bounded random network delay in the NMAS. We achieve this by adding delay measurement signal generator within each agent control system. In this work, the network delay is considered to be half of the measured total delay in the network loop, which can be measured using a ramp signal. This method assumes that the duration for each agent to receive data from its neighbouring agent is equal to the time for the agent’s own transmitted data to be received by its neighbouring agent(s). In the final case, we propose a novel strategy for combining the predictive control with a new gain error ratio (GER) formula. This strategy is not only capable of compensating for a large number of consecutive data losses (CDLs) in the external consensus problem; it can also compensate for network constraints without affecting the consensus convergence time of the whole system. Thus, this strategy is not only able to solve the external consensus problem but is also robust to the number of CDL occurrences in NMAS. In each case, the designed control algorithm is compared with a Proportional-Integral (PI) controller. The evaluation of the NMAS output performance is conducted for each by simulations, analytical calculations, and practical experiments. In this thesis, the research work is accomplished through the integration of basic blocks and a bespoke Networked Control toolbox in MATLAB Simulink, together with NetController hardware.