Studies On Hybrid Ant Colony Optimization And Its Application In Optimization Problems

With nearly 20 years development, ant colony algorithm has successfully applied to the discrete combinatorial optimization, the continuous function optimization and clustering. However, there are the following four deficiencies for ant colony algorithm:(1) The computation complexity for ant colony optimization is too high when Bayesian network structure is learned. And the running time is too slow.(2) The effect is poor when the continuous optimization problems are solved by ant colony optimization.(3) The effect is poor when constrained optimization problems are solved by ant colony optimization.(4) The clustering effects are poor when ant colony clustering algorithm is applied to a number of benchmark data sets.In order to overcome the above four shortcomings, the following improvements for ant colony optimization are provided.Firstly, the constrained ant colony optimization is proposed for the discrete combinatorial optimization problem-learning Bayesian network structure. The add-edge-rule is designed base on the local consistency score criterion in the new algorithm. And the rule is embedded in the framework of ant colony algorithm. Therefore, the new algorithm can use the heuristic to dynamically reduce the search space and reduce the running time during the search process. The empirical tests show that without loss of results accuracy, the convergence speed of the proposed algorithm is significantly 40% faster than that of ant colony optimization. In addition, the Bayesian classifier based on constrained ant colony algorithm is proposed for project's risk prediction of real estate project in the stage of decision. The experiments results show that the new Bayesian classifier performs better for the risk prediction and shows the causal relationship between factors and risk.Secondly, a new ant colony optimization with high performance is proposed for continuous function optimization.In the new algorithm. an adaptive setting mechanism for parameters is designed and a local search operator inspired by artificial bee colony is embedded in ant colony optimization. The simulation results show that the new algorithm can enhance the local search ability of ant colony optimization.Thirdly, a novel diversity differential evolution algorithm with composite trial vector generation strategies, which is inspired by the transition rule of ant colony optimization, is proposed for constrained optimization problems. Inspiring by the transition rule of ant colony optimization, a novel diversity mechanism was designed to utilize the information in infeasible solutions. The experimental results on 13 benchmarks indicate that the new algorithm exhibits obvious superiority in solving the most difficult optimization problems g02,g10 and g13.Fourthly, kernel ant colony clustering algorithm is proposed for cluster problems. The radius kernel function is applied to similarity function to improve the clustering effects. Simulation results show that the kernel ant colony clustering algorithm performs successfully for IRIS dataset. In addition, kernel ant colony clustering algorithm is applied to project's risk evaluation of real estate project in the stage of decision. An example is used to illustrate the whole process. Experimental results show that ant colony clustering algorithm has good prediction effect.