A networked distributed system (NDS) is a system consisting of a collection of spatially
distributed autonomous nodes connected by networked communication to produce
an integrated and consistent behaviour or task. The networked distributed systems
(NDSs) can enable the agents to cooperate and coordinate their activities more effectively
and efficiently. The typical kinds of NDSs, such as complex dynamical networks
(CDNs), multi-agents systems and sensor networks, are ubiquitous in our real world.
Based on the networked communication, many kinds of collective behaviour and applications
of NDSs can be realized, and have attracted increasing attention in past
decades. In particular, synchronization problem in CDNs and target tracking in sensor
networks have been intensively studied in recent years.
It is known that the communication among the agents of NDSs may not be perfect
due to all kinds of environmental constraints. Some communication constraints have
been investigated, such as time delay and communication noise. However, there are
still some important problems for communication constraints, which have been seldom
considered.
(i) In many real-world NDSs, brain networks for example, the communication between
any two nodes consists of multi channels, which independently transmit
the information of the nodes. In existing results, the working states of the channels
are assumed to be the same: all of them are connected or disconnected.
However, in many cases, only part of these channels can work properly, while
other channels malfunction. Such communication constraint results in partial information
of the nodes transmitted. A natural problem is how the NDS behaves
under partial information transmission.
(ii) In electronic systems, the sampling-hold technique is widely employed. The
sampled data can effectively reduce the blocking rate of the communication
medium. However, only “old” information is available for the nodes in the NDS
during each sampling interval. Although some results have been given for NDSs
with sampled data, the derived conditions need to verify some linear matrix inequalities
(LMIs) with a large dimension. Is it possible to propose more easily
verifiable results for NDSs with sampled data?
(iii) The data packet dropout, which is mainly caused by a network-induced traffic
jam, commonly occurs in many NDSs. It is challenging to consider the data
dropout with other communication constraints simultaneously. For example, if
the sampled data were missing, the node cannot update the information of adjacent
node during a whole sampling interval. Thus, how does one analyze the
NDSs with hybrid communication constraints?
This thesis presents our research and developments on the above concerned questions
of NDSs with communication constraints. We mainly focus on three kinds of
communication constraints: partial information transmission, sampled data and data
packet dropouts. Due to the existence of communication constraints, less information
can be used for the agents of the NDS, which leads to more difficulty in analyzing
the dynamical behaviour of the system. Specifically, the following issues will be addressed:
(a) synchronization criteria for CDNs with diagonal or upper triangle inner
coupling matrices; (b) partial synchronization of stochastic differential networks (SDNs)
with switching communication constraint; (c) pinning synchronization of CDNs
with partial discrete-time couplings; (d) targets tracking problem in sensor network
with partial information transmission; (e) distributed filtering design in sensor network
with communication constraints.
For problem (a), the model of CDNs with partial information transmission is constructed,
in which the channel matrix is introduced to indicate the working state of the
channels of each connection. Furthermore, to reflect a more realistic situation, CDNs
with two typical kinds of inner coupling matrices: diagonal matrix and lower triangle matrix are considered. By using the regrouping method, some synchronization criteria
are then derived for the CDNs with partial information transmission.
For problem (b), the switching channel matrices is studied for partial synchronization
of SDNs. The switching law is governed by a Markovian chain. By changing
the active channel number of each switching mode, the influence of communication
constraints on the synchronization process can be discussed in detail.
For problem (c), two communication constraints: partial information transmission
and sampled data are simultaneously considered in CDNs with pinning control. The
sampling protocol is used in the communication among nodes and control signal. Synchronous
and asynchronous sampling protocols are respectively considered. For both
sampling protocols, some easily verifiable conditions are obtained.
For problem (d), the partial information transmission is investigated in sensor networks.
The sensor network is implemented to track two targets, where each sensor can
only monitor a part of states of the targets. The sufficient condition is obtained to guarantee
that the two targets can be tracked by the sensor network with partial information
transmission. This result can be extended to a multi-targets tracking problem.
For problem (e), the distributed filter is designed for the sensor network with hybrid
communication constraints: sampled data and data packet dropouts. The sufficient
condition for distributed filtering is obtained such that (i) without noise, the plant can
be estimated by the sensor network; (ii) for any nonzero L2 noise, the given H1 performance
can be satisfied.
Overall, the main contribution of this thesis can be listed as follows.
● The concept of partial information transmission is first proposed. Each connection
between any two nodes is divided into multi-channels. Unlike the existing
work, the working states of these channels are not the same, which leads to being
able to transmit only partial information of one node. To reflect the activity of
channels, a channel matrix is built for each connection. Furthermore, a regrouping
method is introduced to overcome the difficulty caused by the constraint.
● Complete synchronization, partial synchronization and pinning synchronization are respectively investigated in CDNs with communication constraints. Some
efficient synchronization criteria are derived, which are in terms of LMIs with
relatively smaller dimension. Thus, our results are easily testified, especially for
the large-scale NDSs.
● By constructing different channel matrices in the CDN, the synchronization time
is computed to find out the influence of partial information transmission on the
synchronization process. From the numerical simulations, it can be concluded
that the number and allocation of active channels are two critical factors. Furthermore,
when the channel matrices suffer from abrupt switching, the switching
law has great impact on the synchronization result.
● Although all of the above three communication constraints would lead to less
information of nodes being available, their mechanisms are different. It is practical
to consider them simultaneously in NDSs and the mathematical issues are
not trivial to deal with. In our work, it is the first time to investigate the NDSs
with hybrid communication constraints.
● Under the imperfect communication, distributed filter is studied for sensor networks
to multi-targets tracking or plant estimation. The sufficient conditions
concerning the filtering gains are derived and used to design efficient distributed
H1 filter to achieve the following objectives: (i) the filtering error system is exponentially
mean-square stable in the absence of disturbance and noise; (ii) the
prescribed H1 performance constraint is satisfied.
| Date of Award | 15 Feb 2013 |
|---|
| Original language | English |
|---|
| Awarding Institution | - City University of Hong Kong
|
|---|
| Supervisor | Wing Cheong Daniel HO (Supervisor) |
|---|