Inductive fuzzy neural network for multi-input multi-output dynamic systems modelling
This paper presents a systematic inductive fuzzy neural network for multi-input multi-output dynamic systems modelling based on input/output measurements. An inductive learning algorithm is applied to generate the required fuzzy modelling rules from input/output numerical measurements recorded from the dynamic system. Then, a full differentiable fuzzy neural network is developed to construct the dynamic model of the multi-input multi-output system, while back-propagation algorithm or similar techniques can be further applied to tune the network parameters due to the differentiable nature of the developed network.
| Attachment | Size |
|---|---|
| Paper_Presentation_IPROMS.wmv | 4.26 MB |
Oops! Well spotted, Maria!
I suspect the original conclusion was left out due to lack of space.
I shall speak to our colleagues about restoring it.
Ashraf or Eldaw will respond to your questions.
Many thanks for your contribution.
Cheers.
DTP.
Dear Maria,
Thanks for your Questions and interest in the paper. My reply is as follows:
1- The contribution of the paper is mentioned at the end of the Results Section and at the end of the Paper Presentation which are as follows:
- A new systematic technique for modelling and control of multi-input multi-output dynamic systems was presented.
- An inductive learning technique was employed to construct the model structure using input/output measurements resulting in smaller model size.
- A full differentiable neurofuzzy network is constructed to form an adaptive model resulting more adaptable system.
- Gradient descent is successfully applied to tune the neurofuzzy network weights using feedback error learning scheme.
2- The back-propagation algorithm was used to show the adaptability of the network parameters and ability for online tuning for all of its parameters. There is no specific reasons for using it. Any other adaptation technique can be applied to the network.
Once again thanks for your interest in the paper and hope that my answer covers your questions.
All our Regards
A. Fahmy
Hi Prof. Pham & Maria,
A corrected version of the paper which include a Conclusion section is uploaded to the conference.
Regards
Ashraf
Dear authors,
Thanks for your contribution to IPROMS2007. It's an interesting paper to read. The inductive fuzzy neural network developed is very well-explained.
I have two queries:
1. It seems the contribution of this work has not been concluded. Could you outline the advantages of the proposed inductive fuzzy neural network when compared to the previous FNN?
2. The network parameters are adapted using back-propagation learning algorithm in this paper. Why did you choose this algorithm instead of other adaptation algorithms? have you tried other similar algorithms for adaptation?
Cheers.
Maria