A neurofuzzy approach for the anticipatory control of complex systems

Anticipatory control refers to system regulation based on information about anticipated future states. A preview of the future is typically obtained via predictive models and decisions about changes of state are made at the present taking into account the output of such models. Significant improvements in soft computing methodologies support the development of anticipatory control that integrates planning and control sequencing functions with feedback control algorithms. We present a neurofuzzy approach for anticipatory control using radial basis neural models and fuzzy rules and demonstrate it through nuclear reactor regulation. The control method does not require knowledge of plant parameters or structure. It is model-independent, and thus may be applied to other nonlinear time-varying dynamic systems. Simulation results show that the neurofuzzy anticipatory control approach improves tracking performance and smoothness and may be quite insensitive to noise. The results suggest that it is a flexible and powerful approach that can easily be extended to other problems, and it is compatible with other techniques. The relevance of the approach to the control of large complex systems is also discussed