This letter presents a data-driven solution to the discrete-time infinite horizon linear quadratic regulator (LQR) problem. The state feedback gain is computed directly from a batch of input and state data collected from the plant. Simulation examples illustrate the convergence of the proposed solution to the optimal LQR gain as the number of Markov parameters tends to infinity. Experiments in an uninterruptible power supply are presented, which demonstrate the practical applicability of the design methodology.
30 March 2021 in IEEE Transactions on Industrial Electronics
Uninterruptible power supply (UPS) is an electronic power device that delivers voltage to critical loads and whose application must satisfy standardized performance requirements. Usual control techniques demand the UPS model. In this article, we propose a simple, model-free procedure based on the generalized forced oscillation method to design the control system of a UPS output stage. The procedure involves the design of an inner current loop with a proportional gain and a proportional-multiresonant controller in the output voltage loop. We also compare the closed-loop results obtained in a single-phase 3.5 kVA UPS with the IEC 62040-3 performance criteria and show that the results are equivalent to model-based approaches.
The choice of the reference model is the main task to be performed by the designer in a model reference control design. A poor choice of the reference model may result in a closed-loop performance that bears no resemblance to the one specified and may even be unstable. In this paper, we discuss this issue in the control of multiple-input multiple-output plants in the very typical case where a no plant model is available and the controller structure is defined a priori. Experimental results in a three-tank level control plant show how a naive choice of the reference model leads to very poor closed-loop performance. The problem is then analyzed, exposing the naivete of this design example and deriving the general guidelines for the effective choice of the reference model in multivariable systems. The application of these guidelines to the three-tank plant illustrates their effectiveness.
When applying model reference control to a non-minimum phase (NMP) plant, it is important to include the NMP (transmission) zero(s) in the reference model, otherwise the controller will tend to cancel out these zeros, often causing loss of internal stability. In data-driven (DD) control it is not possible to conceive a priori a reference model with the NMP zero(s) because no model of the plant is available; accordingly, DD design methods tend to fail in NMP plants. For SISO (single-input-single-output) plants, this problem has been solved in [1] by using a flexible reference model in a DD design. This paper presents an extension of that method for MIMO (multi-input-multi-output) plants.