Logistics Distribution Routes Optimization Based On The Physarum Network And Ant Colony

Logistics has been promoted as "the Third Profit Source", and its level of modernization is a significant symbol reflecting the degree of a country’s modernization and comprehensive national strength. With online penetration increasing, postal business, e-commerce and logistics business has a faster growth. Our country is stepping into an e-commerce age, and the logistics business would have a great promoting in the future.Logistics system let logistics have an integrity and systematic planning and development. The goal of logistics system is to get the minimum cost, the best customer services and the highest economic benefit of society in a certain condition. However, the performance and the cost of logistics are related with the structure of the distribution network, distribution route and the distribution cost. Therefore, we need to plan and optimize a whole logistics distribution network to improve logistics efficiency and reduce logistics cost. The logistics distribution network planning has four steps. The first step is optimizing the structure of distribution network. The second is siting facility locations. The third is to optimize distribution routes and the last is conveyance optimization. Among these steps, the third one is the most important and also a difficulty and hot area of research.There are four types of distribution routes optimization. One is transportation problem between two dots, also called as shortest path problem. One is transportation problem between multi-dots, also named as dummy balanced transportation problem. One is transportation with single circuit, also called as travelling salesman problem (TSP). The last one is transportation with multi-circuits, also named as vehicle routing problem (VRP). Among these problems, TSP and VRP are N-P problems. Solutions for these problems are different, and mostly divided into two parts:accurate solutions and approximate solutions. Although accurate solutions can solve small-scale problems efficiently, it takes geometrical time for solving large-scale problems in accurate solutions. In this situation, approximate solutions will be chosen as the best solutions. Most of the approximate solutions are heuristic algorithms. In this paper, author has proposed a new intelligent bio-inspired algorithm, which is a hybrid algorithm with ant colony optimization algorithm (ACO) and a new bio-inspired algorithm proposed by Tero et al.ACO, a heuristic solution, is frequently used for solving combination optimization problems. It has fast solution speed and great robustness, but it also has shortcomings of stagnation behavior and the local optimal solution that comes from adding extra pheromones of local best solutions.Nakagaki, Tero et al. have done many experiments with a kind of simple organism, named as physarum polycephalum. They have found that this creature could form a tubular network when foraging for food. This network has been proved having advantages of lower cost, higher transmission efficiency and greater fault tolerance. Then, Tero et al. have proposed a mathematic model for describing the evolution of the physarum network, which is also called as the physarum network with single pair of inlet and outlet (PNss) in this paper. PNss can be used for solving the single-solution shortest path problems or the multi-solutions shortest path problems. Scientists have spent lots of time for optimizing or designing networks. Nevertheless, there are fewer researches about how to solve combination optimization problems by using this network model. This paper has-proposed two new bio-inspired algorithms by utilizing the feature of the important pipelines being reserved with the evolution of network and combing with ant colony optimization algorithm. One is an ant colony system based on multi-pairs of inlets and outlets (PNMM), donated as PNMM-ACO, which is used for solving TSP in our paper. And the other is an ant colony system based on single inlet and multi-outlets (PNsm), donated as PNSM-ACO, which is used for solving VRP in this paper.In PNss, single inlet and single outlet are selected to calculate the fluxes of the pipelines of network at the first stage and shortest path or paths can be gotten at the end. However, in PNMM, multi-inlets and multi-outlets are selected to calculate the average fluxes at the first stage, and a final network can be gotten at the end. PNmm also shows the feature of the important pipelines being reserved with the iteration. By using this feature, this paper has combined PNmm with ACO and proposed PNMM-ACO for solving TSP. In PNMM-ACO, not only does the ant release pheromone trails, but also does pheromone flow in pipelines. Meanwhile, this paper has changed the approach of updating global pheromone to improve convergent performance. Some experiments in synthetic and benchmark networks show that PNmm-ACO is more efficient than ACO in the field of computational efficiency and robustness for solving medium and small scale TSP. However, this algorithm is further influenced by the initial pheromone which is determined by the initial total pheromone trails carried by an ant. In PNSM. select multi-inlets and multi-outlets are selected to calculate the average fluxes at the first stage, and also a final network can be gotten at the end. This paper has proposed PNSM-ACO for solving VRP by utilizing the same feature of the important pipelines being reserved with the iteration. The difference between PNSM-ACO and PNMM-ACO is that in PNSM-ACO, all of the ants leave from one warehouse (inlet), and the initial pheromones are related with the requirements of the other cities. PNSM-ACO use the similar approach to update the global pheromone trails. However, the results of PNSM-ACO have lower influences with the initial pheromones than the results of PNmm-ACO. Some experiments also show that PNSM-ACO is more efficient than ACO in the field of computational efficiency and robustness for solving medium and small scale VRPsAlthough PNMM-ACO and PNSM-ACO relieve the shortcomings of local optimal solution of ACO, when the scale of problems is enlarged, both two algorithms need more computation cost and time for solving problems so that the algorithms are not applicable in this situation.