Researches On Models Of Physarum Polycephalum Intelligence And Applications

The study of biological behavior can promote new calculation methods and development of artificial intelligence. Over the last decade, a multinucleated single-cell organism, called Physarum polycephalum, has aroused a great interest. The life cycle of Physarum polycephalum is complex. During the vegetative stage, the plasmodium of Physarum polycephalum is able to change shape, and construct a protoplasmic pipeline network of the food sources. In2000, Nakagaki found that the protoplasm pipeline of the slime mould is able to evolve to obtain the shortest pipeline to connect the entrance and exit of a maze. Since then, the intelligence of the low level organism gradually catches researchers’attentions. From different angles, scientists use a variety of methods to research on the Physarum intelligent behavior, trying to explain the internal mechanism and apply it to solve practical problems. This thesis analyzes three main methods among them:1) Tero et al. have built a model for explaining how Physarum solving the shortest path problem, which is based on Kirchhoff law and Hagen Poiseuille law to establish the feedback between pipeline flow and conductivity.2) Adamatzky has used reaction-diffusion process to analog the growth process of Physarum. The network constructing and maze solving of Physarum are studied by utilizing the variables Oregonator model.3) Jones has used a bottom-up modeling approach to build multi-agent model, the model can approximate the Physarum network evolution. Based on the study of the three models, the thesis accomplishes the following three tasks:(1)The current control on resistance in the memristor model, found by HP Lab in2008, inspires us to understand the similarity to the relationship between flow and conductivity in Tero’s model. Simscape and Simulink are utilized to achieve and study the memristor model. Then, a new model based the memristor with some characteristics of superconductivity is obtained. The corresponding simulating element of the new model is established by Simscape and Simulink, as well as the Tero’s model. Experimentally, it is demonstrated that the new model has the function of path selection like the Tero’s model. Moreover, the new model has the feature of adjustable selection speed and better performance in time consume by parameter setting. Finally, a circuit system, built by the simulation elements of the new model, is applied successfully to simulate Physarum to solve maze by finding the shortest path.(2)Through simulating the distribution of food sources concentration on the nutrient-poor substrate, we establish a mathematical model based the concentration gradient to simulate the foraging behavior of Physarum on the nutrient-poor substrate. The food source is abstracted as field source forming the chemical particle concentration spatially. The field strength of one point in the space is related to its Euclidean distance from a food source. A growth point moves along the gradient, and its track shows a spanning tree of all field sources. By comparing the results, it is found that the spanning tree constructed by our model is similar to the tree-like network of Physarum on nutrient-poor substrate. Furthermore, the proposed model can imitate Physarum to construct a spanning tree with pipe branches or to avoid the repellent and find other food sources. When the proposed model is applied to build a Euclidean spanning tree, its topology is almost the same as the corresponding Euclidean minimum spanning tree.(3)After studying the characteristics of the Jones model, a Multi-Agent particle model with evolution mechanism is proposed. Agent in the Multi-agent model of this thesis has fewer sensors which can identify both nutrient source and trail, and is armed with a Motion Counter which controls its reproduction and elimination. Thus the proposed model can imitate Physarum to build various formations of network for the same arrangement of food sources and can adaptively adjust the population of Agents. By increasing agent types and transition rules, the proposed model can simulate Physarum to solve maze.