Research Of Fireworks Algorithm And Its Application In Clustering

Fireworks algorithm is an abiotic swarm intelligence optimization algorithm.This algorithm simulates the explosion behavior of fireworks in the night sky and proposes an explosion search method.The purpose is to solve global optimal solutions to complex problems.The traditional fireworks algorithm has a good optimization ability to optimize the optimal function near the origin,but for those objective functions whose optimal value is not near the origin,the algorithm does not perform well,because the traditional fireworks algorithm adopts a mapping method which is easy to map the cross-border sparks near the origin of the search space.Due to the influence of such human factors,the algorithm convergence of the optimization problem with the optimal value near the origin is accelerated.However,this is not the performance of swarm intelligence.In the Enhanced Fireworks Algorithm(EFWA),a random mapping rule is proposed for this problem.The out-of-range sparks are randomly mapped to arbitrary positions in the feasible region.However,this random mapping rule cannot dynamically perform the mapping position according to the current population evaluation status.Adjustment.For this reason,a new mapping rule is proposed in this thesis.According to the current evaluation progress of the algorithm,the out-of-boundary sparks are dynamically mapped to the nearest fireworks with the best fitness value,and a dynamic mapping fireworks algorithm based on optimal sparks is proposed.By using new mapping rules,the convergence speed and accuracy of the algorithm are improved.The minimum explosion radius detection mechanism is used in EFWA to improve the local search ability of the fireworks with the best fitness value.However,the value of the final explosion radius parameter used in this mechanism is determined by the boundary of the objective function.This method is used for different boundary functions in the algorithm.The impact of convergence in the late period is large,and the target function with a large range of boundaries will be slow to converge at the later stage of the algorithm.For this problem,the final blast radius parameter is given a fixed value to avoid adverse effects on the algorithm's late search due to different boundaries,thus improving the convergence speed of the algorithm.Aiming at elite selection strategy in EFWA algorithm,this thesis proposes a new selection strategy—elite selection strategy,which chooses the best spark in the sample space and randomly selects sparks in the remaining samples(For the next generation number of fireworks,together make up the next generation of fireworks population.Through the new selection strategy,the global optimization capability of the algorithm is enhanced and the convergence speed of the algorithm is improved.Another work in this thesis is to transform the traditional clustering problem into an optimization problem,and proposes a clustering analysis algorithm based on the dynamic spark mapping fireworks algorithm(DMFWA).The algorithm converts the clustering process into a search process for finding the best clustering center point,and uses the powerful global search capability of the DMFWA algorithm to find the best clustering center point in the sample space,avoiding the traditional clustering algorithm(eg,K-means)is prone to fall into a local optimum due to improper initialization.The simulation results of 15 test functions show that DMFWA has greatly improved the accuracy and convergence rate of the algorithm compared to EFWA.For the application of DMFWA in cluster analysis,three standard test sample sets are used in this thesis,and four commonly used clustering algorithms(k-means,k-means++,AP and Agglomerative)are compared with the DMFWA clustering algorithm.The results show that the DMFWA algorithm can obtain better clustering quality.