Tasks, deliverables, and milestones

Tasks

Task 3.1: Hierarchical Control

 Partners: TUDO, TUB, UKS, US, UVA (Lead) , UNIVAQ, UNITN, TUDELF

In task 3.1, we are considering system-wide coordination and control of large-scale systems using a multi-level control framework consisting of several layers where at the lowest layer of the control hierarchy local controllers manage fast dynamics in a small region or for a small sub-process, and where higher-level supervisory controllers take care of slower dynamics and coordination over larger regions. So we get a temporal and spatial division of the control. In case the local controllers are not able to appropriately manage their own region or sub-process, they can turn to their neighbours for assistance, or to a higher-level supervisory controller. The supervisors can in their turn ask assistance from their neighbours or from even higher-level supervisors. The basic idea is to solve the problems as much as possible at the level at which they occur or are predicted to occur (local reduction of variance). Within this framework we mainly focus on the higher levels in the control hierarchy and on the following issues:

• - Coordination within and across the control levels in such a way that the behaviour of each single controller contributes to the efficient, reliable, and smooth operation of the total system. Since we focus on the higher layers of the control hierarchy this will include the use of hybrid systems and control methods, in particular propagation of global control goals to local controllers by means of optimization criteria and constraints.

•  - Development of multi-level, multi-scale models that fit the proposed multi-level control approach and of abstraction techniques with guaranteed properties such that a safe operation of the overall system can be assured.

• - Methods for constructing appropriate temporal and spatial divisions of the overall system.

• - Development of efficient algorithms in the light of the trade-off between accuracy and optimality.

• - Handling model uncertainty in higher-level online control and optimization algorithms.

• - Dealing with variable structures, e.g. the shutdown or re-integration of subsystems.

 

Task 3.1: Hierarchical control structure.

Task 3.2: Distributed Control

 Partners: INRIA, ETHZ, TUB, UKS, RUB, US, UVA, UNIVAQ, UNITN, UNICA, UNIPV, TUE, TUDELF (Lead), RUG, KTH

In a distributed control structure, the controllers operate based on locally available partial information and cooperate with others. In order to obtain a desired performance for the global large-scale system, an integrated approach is needed that is able to determine and assess performance limitations and robustness problems arising from the distributed implementation. Distributed MPC has become a very active research area. The current research concerns not only the issues related to the underlying optimization (feasibility, convergence and computational effort), but also closed-loop issues of stability and robustness. One approach to distributed model predictive control (MPC) that needs further investigation is to partition a possibly large model of the overall system into a set of partially overlapping submodels (where “overlap” means that submodels may share some common states, inputs, and outputs) and then design a simple MPC controller for each submodel. By doing so, each MPC controller uses measurements that are obtained both locally and from neighbours, and optimizes not only local command variables, but also the moves that its MPC neighbours may virtually take. In such a way, cooperation towards common performance objectives is pursued. Because of the uncertainty introduced by the modelling phase (some interactions will be inevitably discarded), and also of the mismatch between the “hypothetical” neighbours’ moves and the moves actually commanded by the neighbours, several research issues need to be investigated: Under which conditions the overall closed-loop system is asymptotically stable? How large is the loss of performance compared to a centralized control scheme? For a given degree of cross-coupling among dynamical models, mainly dictated by the amount of state and input information that is going to be communicated, which is the best way of carving out the submodels? Can the decentralization topology be modified during operations by a supervisory algorithm at a higher hierarchical level without affecting closed-loop stability? Distributed optimization algorithms also have to be studied in detail, with important aspects such as information exchange between subsystems, and cooperation between their controllers. The value of information needs to be considered in a distributed setting. Analysis tools are needed for the study of sub-optimality and trade-offs between performance and uncertainty due to incomplete local information. Some specific research topics include:

• - Distributed  constraint fulfilment in large-scale interconnected dynamical systems

•  - Performance analysis of distributed control in light of interconnection structure, uncertainty and network-induced imperfections.

• - Efficient coordination schemes based on dedicated distributed optimization tools for consensus and coordination of localized controllers.

• - Robustness to failures in components and in the communication structure.

• - Dynamic reconfiguration of the overall system due to the disconnection or shut-down and re-integration of subsystems.

 

 Task 3.2: Distributed control structure 

Task 3.3: Price-based Distributed Optimization and Control

 Partners: TUDO (Lead), US, UVA, UNITN, TUE, RUG, KTH, ULUND 

A promising approach to distributed optimization and control is the use of prize-based coordination mechanisms. Such mechanisms have also been proposed and successfully tested in large discrete resource allocation problems and often are labelled as agent-based systems. The underlying idea is to coordinate the use of scarce resources by a bidding procedure where the “agents” offer and buy resources at certain prices that then enter into the local cost functions. The approach has a strong relationship to Lagrangean relaxation of the coupling constraints. Convergence can be improved by modifying the local problems according to the Dantzig-Wolfe scheme. This approach will be applied to the control of future power systems and to large plants in the processing industries that share resources (AD2). The work will extend the foundations of control of distributed systems by price-based coordination mechanisms as well as apply the theory to the case studies described in AD2 in order to compare different approaches, to discover new challenges for research in theories and algorithms and to demonstrate the benefits to industry. Special attention will be paid to the exchange of ideas, algorithms and results between the areas of electric power systems and the processing industry which so far have been dealt with independently.

Task 3.3: Price-based distributed optimization structure

Task 3.4: Identify new interactions between Task 3.1 to Task 3.3 and AD1 and AD3

 Partners: INRETS, TUDO, TUB, UKS, RUB, US, UVA, UNICA, UNIPV, TUDELF 

In task 3.4, it will be investigated which opportunities and challenges for hierarchical, distributed and price-coordinated control result from the application problems in the domains of transportation and biological and medical systems.

Task 3.5: Integration of running projects in FP 7 ICT

 Partners: TUDO, TUB (Lead), UNITN, UNICA, TUE, TUDELF, RUG, KTH 

In consulting with WP9, the participants in this work package will coordinate the efforts in several FP7 STREPs in which they are involved: DISC, EMBOCON, HD-MPC, WIDE, E-Price. Moreover they will reach out to other projects which are not represented in this WP but deal with similar problems, e.g. AEOLOS, CHAT, C4C and to application projects related to AD2, e.g. the FP7 IP F3 (Fast Flexible Future Factory) on new concepts for modular processing plants. 

Deliverables

To read the available deliverables you must be a Level 1 Partner.

To upload a deliverable you must be a WP Leader.

Please log in to have access to deliverables and to download the template.

Milestones

Milestone No.	Milestone Name	Delivery Date	Document
MST5	Report on the interactions between TAi and ADi	M24
MST11	Report on the interactions between TAi and ADi	M36