Modified direct inverse control of CSTR process with disturbances.

• Autorzy: Ho K.C. , Ponnapalli P. V. S. , Thomson M.
• Języki publikacji: EN

Abstrakty

EN

This paper presents the development of a neural network-based direct inverse controller (DIC) for a Continuous Stirred Tank Reactor (CSTR) process. An inverse model of the CSTR process is developed based on a Multiple-type Feedforward Artificial Neural Network (MFANN). Multi-type FANN incorporates one linear node in the hidden layer while the other hidden nodes are typical sigmoidal functions. A Pruning Selection Scheme (PSS), based on the local comparison of the sensitivity values of weight connections, is used to optimize this MFANN structure. This inverse model is then used in a modified version of a direct inverse controller for a CSTR system. The simulation results indicate that the specially pruned MFANN architecture provides better performance in terms of modelling the non-linear process and in controlling it. The modified version of the DIC provides robust control action compared to the original DIC with and without disturbance.

• Wydawca: Oficyna Wydawnicza Politechniki Wrocławskiej
• Czasopismo: Systems Science
• Rocznik: 1998
• Tom: Vol. 24, nr 4
• Strony: 37--52
• Opis fizyczny: Bibliogr. 16 poz., tab. 3, rys. 11, zał1

Twórcy

autor

Ho K.C.

autor

Ponnapalli P. V. S.

autor

Thomson M.

• Department of Electrical and Electronic Engineering, The Manchester Metropolitan University.