Use this URL to cite or link to this record in EThOS:
Title: The application of artificial intelligence techniques to process identification and control
Author: Xia, T. A.
Awarding Body: University of Wales Swansea
Current Institution: Swansea University
Date of Award: 1998
Availability of Full Text:
Access from EThOS:
The application of artificial intelligence technique (viz, neural networks, genetic algorithms and fuzzy logic systems) to process identification and control has been investigated with different systems. Neural networks and fuzzy logic systems are able to learn the dynamics of a process by training with a set of data obtained from that process and, subsequently, are able to provide a good predictive performance. Genetic algorithms and evolution strategies are non-gradient-based search schemes which facilitate non-linear system optimisation. Thus, they can be applied to the process industries in place of more traditional linear modelling and optimisation. Based on the standard backpropagation network (BPN), a new neural network - the extended backpropagation network (ExtBPN) - has been proposed and tested using different SISO, SIMO and MIMO systems. Two main disadvantages of the standard BPN, i.e. a long training time and poor ability to extrapolate outside the range of training data, are overcome to some extent by using the ExtBPN. A unified strategy for model-based predictive and model reference control has been developed based on the optimisation of a cost function which contains the feedback of error information for the adjustment of future set-points in order to compensate for the mismatch between the model and the real process. In this way, the inherent advantages of both feedback and feedforward control have been utilised. Several new control strategies have been developed from this basis and tested with linear or non-linear, SISO or MIMO systems using neural network and fuzzy process models. These new control strategies (viz. the generalised horizon adjusted predictive (GHAP) control), the predictive direct model reference control (PDMRC), the modified predictive internal model reference control (MPIMC) and the predictive generic model reference control (PGMC) in which the cost function is optimised using genetic algorithms and evolution strategies have been applied successfully to different processes. It has been shown, further, that a variety of time-integral performance criteria can be employed for the design of PID and model-based predictive controllers.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available