My research interests currently embrace plug-and-play control, distributed state estimation and identification of cell population models.
Research has been supported by the EU projects HYCON2 and HYGEIA.
Plug-and-Play control of AC and DC islanded microgrids
We study decentralized control schemes for AC and DC Islanded microGrids (ImGs) composed by the interconnection of Distributed Generation Units (DGUs), typically renewables. Local controllers associated to DGUs must regulate voltage (also frequency in the AC case) at the point of common coupling of each DGU and guarantee stability of the overall ImG. Our goal is to develop control design algorithms that are scalable, meaning that the computational complexity for designing a single controller is independent on the size of the whole ImG. Moreover, when a DGU wants to plug in or out (i) the possibility of performing the operation without spoiling voltage stability in the whole ImG is automatically tested by solving an optimization problem on the DGU hardware (ii) if the operation is allowed, only nighboring DGUs have to retune their local controllers. Automatized, safe and reliable plug-in and out operations enable the deployment of ImGs that can grow over time, where the shutdown of replacement of DGUs can be done ina safe way and where meshed interconnection topologies are allowed, so as to achieve robustness against individual faulty lines. To explore this concept, we test of our controllers through extensive simulations in MatLab/Simulink and PSCAD and through real experiments (in cooperation with the Microgrid group at Aalborg University).
Plug-and-Play control, state estimation and fault detection
We consider large-scale systems represented by a coupling graph between subsystems and develop distributed control schemes capable to guarantee asymptotic stability and satisfaction of constraints on system inputs and states. In particular, we focus on design procedures that enable plug-and-play (PnP) operations, meaning that (i) the addition or removal of subsystems triggers the design of local controllers associated to successors to the subsystem only and (ii) the synthesis of a local controller for a subsystem requires information only from predecessors of the subsystem and it can be performed using only local computational resources. We also study how to recast the design of local controllers into optimization problems that can be solved in parallel by smart actuators equipped with computational resources and capable to exchange information with neighboring subsystems. Our PnP design method hinge on the small-gain theorem for networks and on tube-based Model Predictive Control (MPC). Recently, we have extended principles of PnP design to the synthesis of distributed state observers and output-feedback controllers. Furthermore, we have developed PnP fault detection schemes so as to (i) unplug faulty subsystems before the fault gets propagated in the network (ii) safely re-plug in disconnected subsystems once issues have been solved. We applied our method to the frequency control in power networks and the stabilization of large-scale mechanical systems. These examples are included in the Plug and Play Model Predictive Control (PNPMPC) toolbox for MatLab
Distributed State Estimation
Sensor networks are collections of small, low power consuming and possibly cheap sensing devices, with communication and computation capabilities. When used for monitoring large-scale systems it is often impractical to collect and process all measurements in a centralised unit. An alternative is provided by distributed state-estimation schemes where each sensor process measurements only locally and take advantage of communication with neighbouring units for gaining additional information about unobserved dynamics. This has several advantages, including scalability, and decomposition of a big state-estimation problem into small ones that can be solved in parallel. The main challenge is how to distribute computations without spoiling convergence of the estimates to the true state. I am especially interested in distributed moving horizon estimation algorithms, due to their capability of taking into account physical constraints on state and noise variables.
Identification of cell population models
Significant cell-to-cell heterogeneity is ubiquitously observed in isogenic cell populations. Recent works have acknowledged the importance of modeling variability in gene expression, either intrinsic (coming from stochasticity of chemical reactions) or extrinsic (coming from morphological and physiological differences between cells). We describe extrinisic variability using Mixed-Effects models, i.e. ODE models with stochastic parameters accounting for differences among individual cells. Population-level parameter distributions are then identified using data obtained from time-lapse single-cell fluorescent microscopy (see the side picture). An advantage of ME models is that specific parameters are attributed to individual cells. This information about cell identity can be used for (i) learning how environmental cues (such as cell-size and local cell density) affect intracellular processes underlying gene expression (ii) study how cell-specific features are inherited from mother to daughter cells. Besides calibrating ME models to gene-expression data, we studied how to validate them and assessed their advantages over population-averaged models and models tailored to intrinsic variability sources. Performance of ME models has been tested both on synthetic and real data. In particular, in cooperation with INRIA and Universite' Paris Diderot (France) we applied the ME modeling framework to single-cell longitudinal data of the yeast osmoregulation process.