Research On Tabu Search And Its Application In Forward Neural Network

Tabu Search (TS) is a new meta-heuristic algorithm, first proposed by Fred Glover. It is formed a set of algorithms by simulating or showing some natural phenomenon or intelligent procedure as Simulated Annealing (SA), Genetic Algorithm (GA), Ant Colony System (ACS) and Chaos, for the goal to compute the optimization solution of engineering problems. TS has being become the research hot in Computational Intelligence fields, because of its special and flexible memory mechanism and respective tabu criteria to avoid circuit searching, also has being paid wide attention to by many scholars. TS first was applied successfully to Job-shop problems, and then applied to other fields such as combinatorial optimization and function optimization.This paper has mainly finished three research jobs based on existent research results:(1) Studied algorithm performance how effected by selecting parameters, taking TSP as an example, focusing on two parameters:(1)initialing solutions algorithms;(2) intensification and diversification strategy.(2) Studied TS integrated with GA, proposed that take the mutating operation of GA to apply to TS.(3) Studied TS applied to forward neural network, taking TS as BP network's training algorithm.This paper introduced and discussed TS's main principles and some correlative basic concepts, did many simulative experiments, verified the feasibility and correctness of the above job, and obtained some corresponding results:(1) Taking TSP as an example, proposed some constructive advices for the initialing solutions algorithms based on the problem size and application acquirements. Proposed a novel adaptive intensification and diversification strategy for solving the conflict between intensification searching and diversification searching.(2) Taking the probability mutating operation and the adaptive mutatingoperation separately into TS, enhanced TS' optimizing ability, reduced the sensitivity of TS toward initial solutions.(3) Aiming at the BP algorithm's weakness that it is essentially a local optimization algorithm, applying TS into the forward neural network for globally optimizing its weights, enhanced the forward neural network's convergent ratio, convergent precision and convergent speed.At last, summarized the research results and looked forward application foreground for TS.