Research On Fruit Fly Optimization Algorithm And Its Applicaitons

Fruit fly optimization algorithm (FOA) is a new branch of swarmintelligence theory. FOA can provide efficient solutions for optimizationproblems through intelligence which generated from complex activities such ascooperation and competition in fruit flies foraging process. This paper mainlyfocuses on FOA and its applications such as neural network training for patternclassification, the gray neural network parameter optimizing for forecasting andimproving discrete FOA to deal with the permutation flow-shop schedulingproblem.Firstly, fruit fly optimization algorithm is introduced and its developmentsare reviewed. The applications of FOA are summarized.Secondly, FOA is used to optimize parameters of the gray neural networkfor forecasting. This method improves the accuracy of forecasting and speed upthe degree of convergence，and then overcomes drawback of gray neuralnetwork. The algorithm can achieve better prediction effect to the small sampledata and can be widely apply to similar prediction.Thirdly, Structure-improving Fruit Fly Optimization algorithm (SFOA) fortraining artificial neural network (ANN) is proposed. The method of equallength coding based on link structure which map different network structure wasproposed. The strategy of adaptive variable step size in the smell-based searchstage brings about the dynamic balance between global and local optimizingcapability. By training the structure and connection weights of ANNsimultaneously, the proposed algorithm eliminates effectively some redundant structure of ANN, in order to improve the training efficiency and classificationability of the neural network.Fourthly, the multi-population adaptive discrete fruit fly optimizationalgorithm (MADFOA) was proposed to deal with the permutation flow-shopscheduling problem. The limitation of FOA in discrete optimization andcombinatorial optimization is discussed and the corresponding algorithm is putforward. There are three new points in this algorithm. The chaos theory wasintroduced to initialize the path. The strategy of adaptive variable step size anddifferent random search methods for different steps were adopted in thesmell-based search procedure. The migration operation and elite library wereintroduced in the multiple population co-evolutionary mechanism. Finally,simulation results and comparisons based on the benchmark testing setsdemonstrate the effectiveness and robustness of the proposed algorithm.Finally, The next research directions and potential applications areproposed.