Research On Swarm Intelligent Optimization Algorithm Of Glowworm And Its Application

Swarm intelligence optimization algorithm is a new way to solve the optimization problem, which has been widely concerned by scholars at home and abroad. Glowworm Swarm Optimization (GSO) algorithm is a novel swarm intelligence optimization algorithm, which simulates mating and foraging behavior of natural glowworm.Because of merits of its less parameters, simple process, easy implementation and capturing multi minimum points, GSO has been applied widely in many fields such as biology, engineering, management, military and economy, et al. However, it is difficult for GSO algorithm to avoid the inherent defects of swarm intelligence algorithm, such as premature convergence, low precision and poor stability. In addition, there are many discrete combination optimization problems that need to be solved in practical application, so it is urgent to improve the traditional GSO algorithm.Based on the analysis of the traditional GSO algorithm, the algorithm has been improved from the traditional continuous GSO discretization strategy, population initialization and moving step etc. The improved algorithm is applied to feature selection and pattern classification. Finally, the research results are used to solve the problem of agricultural drought evaluation and prediction. The main research work and creative achievements are summarized as follows:(1) Aiming at the problem of combinatorial optimization, the traditional GSO algorithm is not suitable for solving the discrete problem. Based on the analysis of the Binary GSO algorithm, improved binary discretization strategy of GSO algorithms are proposed by two kinds of probability mapping function which are modified Sigmoid function and Gauss mutation function. The convergence of proposed algorithms is analyzed and proved in theory. Then Benchmark standard test functions are employed to test the proposed algorithm. The experimental results show two kinds of improved binary discrete GSO algorithms have good performance on stability,convergence speed and computational accuracy and other advantages comparsing with BGSO algorithm,(2) Feature selection is an important method of data preprocessing for machine learning and pattern recognition. Especially, there are some high dimensional data sets which their computational complexity is so high that they greatly affect the performance of mining algorithm. Therefore,a new feature selection method based on improved binary glowworm swarm optimization algorithm (SB GSO) and theory of fractal dimension is proposed. In this method, fractal dimension is taken as the evaluation criteria for attribution subsets and binary glowworm swarm optimization algorithm as a kind of search strategy. To verify the feasibility and effectiveness of the proposed method, UCI datasets are used in the experiments. Finally, the proposed feature selection method is applied to the agricultural meteorological drought data for building drought evaluation index system.(3) Aiming at defects of slow learning speed, trapped in local solution and inaccurate operating result of the traditional BP neural network with random initial weights and thresholds, a parallel ensemble learning algorithm based on improved binary glowworm swarm optimization algorithm (GBGSO) and BP neural network is proposed,namely GBGSO-BPNN algorithm. Agricultural meteorological data in Northern Anhui for ten years selected as the experiment datum, the experimental results show that the algorithm has obvious advantages over the traditional BPNN algorithm and BGSO-BPNN in terms of convergence speed and operation accuracy. Therefore, GBGSO-BPNN algorithm can effectively improve the accuracy of agricultural drought assessment and prediction, which provides an effective and feasible method for assessment and prediction of agricultural drought level.(4) Beacause of defects of slow convergence speed, trapped in local solution and poor stability for traditional GSO algorithm solving multimodal function,an artificial GSO algorithm based on the theory of good point set (GPSGSO) is proposed. It adopts the powerful local optimization ability of the initial uniform distribution of glowworm generated by the good point set theory and embeds inertia weight function into GSO to dynamically tune moving step length. The convergence of the algorithm is proved theoretically. Experimental results of 16 Benchmark standard functions show that the improved algorithm is superior to traditional GSO algorithm in convergence efficiency,computational precision and running stability.(5) The performance of SVM classification depends on the chosen model,and the choice of the model mainly involves the selection of the kernel function and its parameter. The values of different kernel functions and their parameters have great influence on the classification accuracy. Therefore,a SVM classification method based on GPSGSO algorithm, namely GPSGSO-SVM algorithm is proposed. The GPSGSO algorithm is used to optimize parameter g of the radial basis kernel function and the penalty factor c of SVM, and the optimal parameters are obtained to improve the accuracy and stability of SVM classification. The experimental results show that the GPSGSO-SVM algorithm has the advantages of fast convergence speed, high stability and high accuracy compared with other similar algorithms with UCI data set.