Adaptive Fine-Tuning for Large-Scale Complex Nonlinear Systems

Practical Large-Scale Nonlinear Control Systems (PLSNCS) require an intensive and time-consuming fine-tuning process in order to achieve a satisfactory - or, even, acceptable - performance. In the majority of PLSNCS the fine-tuning process is performed by experienced personnel based on field observations and by experimenting with different combinations of controller parameters, without the use of a systematic approach. Adaptive Optimization (AO) methods such as the SPSA or AFT1 provide probably the most promising approach for the development of a systematic methodology for automatic and efficient fine-tuning of PLSNCS. However, despite the success of AO methodologies in particular fine-tuning control applications, these methods suffer from the serious problem of not guaranteeing efficient transient behaviour due to the use of random perturbations. A new algorithm (AFT2) has been developed and analyzed for alleviating this problem. This paper presents a comparative evaluation of AFT2 (versus SPSA and AFT1) when applied to three different traffic control applications: urban traffic control, network-wide motorway ramp metering and motorway variable speed control. It is demonstrated by means of simulation investigations that AFT2 not only overcomes the aforementioned problem of poor transient performance but also achieves significantly faster convergence than SPSA and AFT1.)