Abstract

This paper discusses the use of GAs (Genetic Algorithms) and TS (Tabu Search) to design NNCs (Neural Network Controllers) for Real-time control of flows in sewerage networks. Genetic algorithms evolve the weights for Neural Networks Controllers. We apply a modified Tabu Search algorithm in a novel fashion, to select the most relevant training data, in order to reduce the training time.

The comparison between this approach and various fixed penstock control settings, and genetically-designed Neural Networks, is discussed. This paper reports experiments demonstrating that GAs are both effective and robust to design Neural Networks controllers in sewerage network control problems. To confirm whether the GA-Tabu training algorithm has statistically significant better performance than other data selecting algorithms, a t-test with a 5% signigicance level is examined. Use of the Tabu algorithm
reduces the training time without affecting the results.