Research On Crowd-oriented Parallel Multi-objective Evacuation Model

During large-scale evacuation process, different locations of the evacuating object create distinct spatial-temporal route selection circumstances and judgments', thus optimizing the allocation of large-scale spatial-temporal routes of respective evacuating objects so as to improve the emergency performance of the overall system, is a crucial scientific problem. Most of the existed evacuation researches model solely on pedestrian or vehicle object and generally focus only on single-objective optimization. They lack the characterization and depiction of evacuation process in the angle of pedestrian spatial-temporal evacuation, which can hardly effectively analyze the whole emergency process from such aspects as the spatial-temporal rationality of routing selection for each evacuating object, the feasibility and performance in the whole system for all selected routes. Consequently, it create an adverse condition to generate a scientific evacuation solution to carry out systematic emergency polices and contingency measures. This dissertation focus on multi-objective evacuation model under emergency situation in large public places, the spatial-temporal transmission law of pedestrian behaviors, traffic features and multi-objective spatial-temporal evacuation routing algorithm in the environment involved buildings and road net. Major research contents in this dissertation include:Taking into account the individuals'psychology of panic and evacuating at the fastest rate under emergency situation in a large stadium, this dissertation, centering on problems of pedestrian evacuation in a large stadium and its surrounding road network, create three evaluation indexes including the shortest evacuation time, the lowest of the maximum crowding degree and the shortest total path length. This dissertation develops a multi-objective evacuation model, and by combined with evolution optimization theory, multi-objective evacuation route assignment optimization algorithm based on HEMO model is also presented. Compared to NSGA2algorithm, the experimental results show that the proposed multi-objective evacuation model and its associated heterogeneous multi-objective evacuation route assignment algorithm can provide a spatial-temporal evacuation solution with better convergence, diversity and lower crowding degree, which is a beneficial attempt for the perfection of the pedestrian evacuation model. To overcome the time efficiency disadvantage of single thread CPU HEMO algorithms in large scale pedestrian evacuation route assignment, a heterogeneous evacuation model, which utilizes CPU to handle complicated logic operation while adopts CUDA GPU as co-processor to handle large amounts of parallel data operation is designed to provide an alternative theoretical analysis method and technology for both route assignment research and evacuation models which based on heterogeneous computing and MOP. Based on HEMO, the multi-population cooperative evolution method offers a new way to improve the diversity of route assignment, and a feasible and efficient implementation for this model is provided by an explicit parallelization of evolutionary operators and its multi-population computing pattern.Considering the shortcomings of present multi-object evolutionary algorithm in large scale stadium evacuation and satisfying the demands of generic abstraction to extract design pattern, a new software engineering design pattern aiming at stadium evacuation and establishing interface for multi-category evolutionary algorithms is proposed in this dissertation. Different from traditional strategy pattern, the proposed pattern defines a common interface for various multi-object route optimizations in stadium evacuation as well as large amounts of data abstraction objects, which provides system abstraction and theory guidance for building evacuation route distribution library. Therefore, this dissertation does a beneficial attempt in setting up and realizing evacuation algorithm abstract rules whilst designing paradigm for practical optimization problem. Algorithm implementation methods under this pattern are also explored.By studying the deficiencies of multi-objective evolution algorithms in addressing stadium evacuation of large-scale pedestrian and the demands of generic programming and extracting design patterns in software engineering, this dissertation create an open-source evolution algorithm library StupidAlgorithm which is based on the proposed design pattern. The proposed library is cross-platform, C++Ox conformant, and provides the ability to automatically generate basic statistical files and plotting outputs of batched experimental results for further optimal parameter combinations comparison and analysis.