Multi-strategy Collaborative Improved Fireworks Algorithm And Its Application Research

Fireworks algorithm(FWA)is a swarm intelligence algorithm proposed by Professor Tan Ying of Peking University in 2010,which simulates fireworks explosion and combines random search with evolutionary computation.The fireworks algorithm is different from other swarm intelligence algorithms.It works with a variety of different operators,realizes information exchange and sharing,and can deal with complex optimization problems.The existence of operators such as explosion,mutation and selection makes the algorithm have broad research prospects.This paper mainly studies the multi-strategy collaborative improved fireworks algorithm and its application.The main work is as follows:(1)In order to solve the defects of slow convergence speed,easy to fall into local optimum and low optimization accuracy of traditional Fireworks Algorithm(FWA),a Multi-strategy Improved Fireworks Optimization Algorithm(MSFWA)is proposed.The performance of the algorithm is improved by dividing the fireworks operator level and performing the nonlinear decrease of the search step size,automatically switching the mutation mode after increasing the mutation operator mutation information,and ’ multi-elite retention + random / elite retention ’ three strategies.Finally,experiments are carried out on 16 benchmark test functions and reducer design engineering constrained optimization problems.The experimental results show that MSFWA has better optimization accuracy and performance than FWA and other heuristic intelligent algorithms.(2)Aiming at the shortcomings of traditional fireworks algorithm(FWA),such as low optimization accuracy and blindness of explosion update,a hybrid fireworks algorithm based on Arithmetic Optimization Algorithm(AOAFWA)is proposed.Firstly,the algorithm uses multiplication and division operations and addition and subtraction operations in arithmetic to realize the explosion operation of fireworks,thereby improving the convergence ability and optimization accuracy of the algorithm.Secondly,the ’ elite ’ selection strategy is used to replace the ’roulette ’ selection strategy to reduce the complexity of the algorithm.Finally,the performance of AOAFWA is tested by optimizing 8 benchmark functions,and the effectiveness of AOAFWA is evaluated by classifying four UCI basic data sets with spectral clustering algorithm.The experimental results show that AOAFWA can improve the accuracy of function optimization by 200 orders of magnitude on four test functions.In the application of spectral clustering algorithm,the average fitness value is at least 3.3 % better than the original algorithm,and the overall performance is excellent.(3)As a supervised learning method in the field of machine learning,support vector machine(SVM)is often used in classification experiments of highdimensional data and small samples.However,the feature subset of the sample and the selection of SVM penalty parameters and kernel parameters often affect the classification performance.Therefore,this paper uses the above AOAFWA algorithm to collaboratively optimize sample features and SVM parameters to improve sample classification performance.The feasibility of the scheme is verified by 8 UCI and breast cancer classification data sets as test sets,and compared with traditional fireworks algorithm and classical genetic algorithm,particle swarm optimization algorithm and grid search algorithm.The experimental results show that the algorithm can obtain relatively high classification accuracy with fewer sample features.