| EPSRC Reference: |
GR/K08000/01 |
| Title: |
RECURRENT NEURAL NETWORKS AND THE PROBLEM OF INVERSE MODEL CONTROL. |
| Principal Investigator: |
Kambhampati, Dr C |
| Other Investigators: |
|
| Researcher Co-Investigators: |
|
| Project Partners: |
|
| Department: |
Cybernetics |
| Organisation: |
University of Reading |
| Scheme: |
Standard Research (Pre-FEC) |
| Starts: |
01 October 1994 |
Ends: |
30 September 1995 |
Value (£): |
35,833
|
| EPSRC Research Topic Classifications: |
|
| EPSRC Industrial Sector Classifications: |
|
| Related Grants: |
|
| Panel History: |
|
|
Summary on Grant Application Form |
|
The investigation of recurrent neural networks embedded within a differential geometric framework. The use of established non-linear system analysis techniques in order to derive the inverses of trained recurrent networks. The preliminary study of internal model control (IMC) strategies utilising neural network models. Progress:The initial period of this pilot study centred on the familiarisation with fundamental concepts of neural networks and the state space representation of systems. One recurrent network architecture, the Hopfield Network, has been expressed in terms of state space equations to allow analysis using Lie algebra. An analysis relating the invariant characteristic of the relevant order of the recurrent network to the topology was carried out. To further enable a better understanding of these fundamental properties, empirical] testing of recurrent networks as both representations of the systems and inverse systems is being carried out on simulations as well as real plants. The former being a Simulated Continuous Stirred Tank Reactor, the latter being a Cereal Extruder. At the same time IMC and network based control strategies are being implemented on simulated plants. Further work is being undertaken to refine the networked models of the two plants mentioned before.One of the important features of the work has been the corroboration of the observation that the relative order of the network model is dependent upon the topology of the network. At the same time, these tests further confirm that the network is capable of identifying the relative order of the system. This is important when the relative order of the system is not known, or when the model obtained from first principles is a poor approximation of the real system.The research programme is on schedule, and is meeting the relevant milestones. This can be seen in the progress made so far, and also from the publications.MeetingsThe research group, consisting of the RA, and the investigators, has had two meetings, the first at Reading and the second at Newcastle. At the same time there has been regular contact through e-mail and the talk facility over internet. A third project meeting is scheduled for March 1995.Publications1) Relative Order and Inverses of Recurrent Networks - Paper has been submitted to Automatica for a second round of reviews. 2) Relative order defines a topology for recurrent networks to be presented at IEE, International Conference ANN95, Cambridge, U.K. 1995.
|
|
Key Findings |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Potential use in non-academic contexts |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Impacts |
|
| Description |
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk |
| Summary |
|
| Date Materialised |
|
|
|
|
Sectors submitted by the Researcher |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
| Project URL: |
|
| Further Information: |
|
| Organisation Website: |
http://www.rdg.ac.uk |