MASSIVELY-PARALLEL ANALOG TABU SEARCH USING NEURAL NETWORKS APPLIED TO SIMPLE PLANT LOCATION-PROBLEMS

Neural networks and tabu search are two very significant techniques wh ich have emerged recently for the solution of discrete optimization pr oblems. Neural networks possess the desirable quality of implementabil ity in massively parallel hardware while the tabu search metaheuristic shows great promise as a powerful global search method. Tabu Neural N etwork (TANN) integrates an analog version of the short term memory co mponent of tabu search with neural networks to generate a massively pa rallel, analog global search strategy that is hardware implementable. In TANN, both the choice of the element to enter the tabu list as well as the maintenance of the decision elements in tabu status is accompl ished via neuronal activities. In this paper we apply TANN to the simp le plant location problem. Comparisons with the Hopfield-Tank network show an average improvement of about 85% in the quality of solutions o btained.