Research On Multi-subpopulation Artificial Bee Colony Algorithm With Memory And Simulation Of Crowd Evacuation

Artificial Bee Colony(ABC)algorithm is a swarm intelligence algorithm that imitates the cooperative foraging behavior of bees.Compared with other swarm intelligence algorithms,ABC is more simplicity and flexibility,and has fewer control parameters.Since it was proposed,ABC has aroused wide attention from researchers,and its application field has also expanded from simple numerical optimization to neural network training,filter design and other engineering fields,and has achieved very good application results.However,ABC algorithm still has some shortcomings: First,the population of ABC algorithm is single,and it is difficult to achieve a full search in the global scope.Second,ABC algorithm search is blind.In the stage of neighbor nectar source selection and location update,random selection and update are used,which has great uncertainty.In addition,ABC algorithm uses a single search strategy,which leads to strong exploration capabilities and weak development capabilities.A single search strategy is also difficult to adapt to the search requirements of different problem areas.The above problems affect the solution performance of ABC and restrict its further development.Some improvement methods have been proposed,but few studies have considered the influence of search experience on the performance of ABC algorithm.Studies have shown that bees have memory and learning behaviors when collecting nectar,which can help bees find a source of high-quality nectar that can be mined faster.Therefore,by adding memory and learning behavior to the search of artificial bees,using experience to guide the search of the colony is of great significance for improving the performance of ABC algorithm.Aiming at the existing problems,this thesis proposes two improved ABC algorithms.In these two algorithms,a full search in the global scope is achieved by dividing sub-populations.the first improved algorithm uses a memory table to store excellent nectar source selection experience,and the second algorithm extends the experience to the selection of search strategies,using Q-table establishes an adaptive selection mechanism of search strategy.Finally,the two improved ABC algorithms were applied to the path planning problem in crowd evacuation,and the ability of the two improved algorithms to solve practical problems was verified.The main work and innovations studied in this thesis include:(1)Propose a multi-subpopulation artificial bee colony algorithm with memory table.In this algorithm,the population is firstly divided,and the algorithm is prevented from falling into the local optimum through collaboration among multiple sub-populations.Furthermore,a memory table is introduced into each subpopulation to store the excellent nectar sources used when the update is successful,which realizes the search under the guidance of experience and avoids the blindness of the search.At the same time,an adaptive search step improvement strategy for employed bees and onlooker bees is established to overcome the randomness and uncertainty of search.(2)A multi-subpopulation artificial bee colony algorithm with Q table is proposed.Based on the division of sub-populations,this algorithm combines the idea of Q-learning to construct the learning and memory behaviors of search strategies within sub-populations,and further extends the search experience to the selection of search strategies.By introducing a Q table in each subpopulation,an adaptive learning and selection mechanism of multiple search strategies is established to choose the right search strategies for different bees to balance exploration and development.In addition,the neighborhood range of the onlooker bee and the search direction of the scout bee are improved and adjusted to further improve the algorithm development capabilities.(3)Two improved ABC algorithms are applied to the path planning problem in crowd evacuation.The nectar source in ABC corresponds to the actual evacuation exit,and the employed bee corresponds to the leader in the crowd.The algorithm is used to search in the evacuation scene to help the crowd plan reasonable evacuation routes,thereby improving the evacuation efficiency and shortening the evacuation time.In summary,this article aims at the problems of ABC algorithm,starting with the memory and learning behavior of bees in nature,and proposes two improved ABC algorithms.The first algorithm introduces neighbor nectar selection experience in ABC algorithm,and the second algorithm establishes an adaptive selection mechanism of search strategy.Comparison experiments with other improved ABC algorithms show that the two improved ABC algorithms proposed in this thesis have better search ability and convergence speed.Finally,the two improved ABC algorithms are used to solve the path planning problem in crowd evacuation.Experiments show that the path planning method based on improved ABC algorithms effectively improves the evacuation efficiency,and shortens the evacuation time.