In this paper we hybridize ant colony optimiza- tion （ACt） and river formation dynamics （RFD）, two related swarm intelligence methods. In ACt, ants form paths （prob- lem solutions） by following each other＇s phe...In this paper we hybridize ant colony optimiza- tion （ACt） and river formation dynamics （RFD）, two related swarm intelligence methods. In ACt, ants form paths （prob- lem solutions） by following each other＇s pheromone trails and reinforcing trails at best paths until eventually a single path is followed. On the other hand, RFD is based on copy- ing how drops form rivers by eroding the ground and de- positing sediments. In a rough sense, RFD can be seen as a gradient-oriented version of ACt. Several previous experi- ments have shown that the gradient orientation of RFD makes this method solve problems in a different way as ACt. In particular, RFD typically performs deeper searches, which in turn makes it find worse solutions than ACt in the first exe- cution steps in general, though RFD solutions surpass ACt solutions after some more time passes. In this paper we try to get the best features of both worlds by hybridizing RFD and ACt. We use a kind of ant-drop hybrid and consider both pheromone trails and altitudes in the environment. We apply the hybrid method, as well as ACt and RFD, to solve two NP-hard problems where ACt and RFD fit in a different manner： the traveling salesman problem （TSP） and the prob- lem of the minimum distances tree in a variable-cost graph （MDV）. We compare the results of each method and we an- alyze the advantages of using the hybrid approach in each case.展开更多
文摘In this paper we hybridize ant colony optimiza- tion （ACt） and river formation dynamics （RFD）, two related swarm intelligence methods. In ACt, ants form paths （prob- lem solutions） by following each other＇s pheromone trails and reinforcing trails at best paths until eventually a single path is followed. On the other hand, RFD is based on copy- ing how drops form rivers by eroding the ground and de- positing sediments. In a rough sense, RFD can be seen as a gradient-oriented version of ACt. Several previous experi- ments have shown that the gradient orientation of RFD makes this method solve problems in a different way as ACt. In particular, RFD typically performs deeper searches, which in turn makes it find worse solutions than ACt in the first exe- cution steps in general, though RFD solutions surpass ACt solutions after some more time passes. In this paper we try to get the best features of both worlds by hybridizing RFD and ACt. We use a kind of ant-drop hybrid and consider both pheromone trails and altitudes in the environment. We apply the hybrid method, as well as ACt and RFD, to solve two NP-hard problems where ACt and RFD fit in a different manner： the traveling salesman problem （TSP） and the prob- lem of the minimum distances tree in a variable-cost graph （MDV）. We compare the results of each method and we an- alyze the advantages of using the hybrid approach in each case.
