Research On The Modified Artificial Fish Swarm Intelligent Optimization Algorithm And Its Application

Artifical fish swarm algorithm(AFSA) is an emerging meta-heuristic bionic swarm intelligence optimization algorithm, which simulates the interactive social behaviors of fish populations to achieve the swarm intelligence.Based on the introduction of the development process and study status quo of the hot optimization technologies, the status of artificial fish swarm algorithm was discussed. Based on analyzed AFSA models and artificial fish behaviors, the impact of the AFSA main parameters was analyzed, and influence of AFSA algorithm parameters and parameter settings general guiding principles were obtained. Under the theory of intelligent algorithms unified framework, AFSA theory was described by using the unified framework, and the convergence was analyzed.For the contradiction of AFSA view and step with different requirement at different stages of the algorithm, a segmented adaptive function coefficient was designed to improve artificial fish view and step adaptively. Three segmented adaptive function of power function, linear function and exponential function were designed as artificial fish view and step segmented adaptive coefficients. Power function could make view and step decay quickly with the fastest rate of decay, which was mainly used for the problems without prominent local optimal optimization. The slowest decay function, linear function decayed with uniform rate, which was mainly used for the problems with prominent local optimal optimization. The decay effect of exponential function was between power function and linear function. With the segmented adaptive function coefficient, AFSA parameter robustness has been improved greatly, and the algorithm complexity has been improved.For the problem with imprecise optimal solution, artificial fish individuals scattered and reduced convergence efficiency of basic artificial fish swarm algorithm in the late, the AFSA based on evolutionary strategy was proposed according to evolutionary theory. The elimination and clone mechanism by simulating asexual reproduction(ECM) and the reorganization method with adjustable weights simulating sexual reproduction(SR) were proposed respectively to improve the AFAS according to simulating the biological evolution of asexual and sexual reproduction mode. The elimination and clone mechanism cloned the high fitness individuals instead the low fitness ones, which improved the inclusive fitness of artificial fish populations. The reorganization method with adjustable weights was proposed to realize the sexual reproduction of artificial fish individual. The progeny fish was generated by the method of recombination with adjustable weights based on evolutionary strategy. The parent individuals selected uncertainty of evolutionary strategy made the progeny fish keep more features. The Evolutionary Strategy not only improved the fitness of overall fish populations but also kept the population diversity, which effect was better than the elimination and clone mechanism.For the limited effect with single improved method, hybrid AFSA had been studied. A new jump behavior of the artificial fish had been proposed to expand the artificial fish behaviors, which overcame the problem of local minima outstanding. Segmented adaptive function method and the elimination and clone mechanism were coalesced to be the hybrid algorithm, an adaptive AFSA with elimination and clone mechanism(ECMA). Given the limitations of the elimination and clone mechanism, the hybrid segmented adaptive AFSA based on sexual reproduction(SRA) had been studied, which also has good performance. The hybrid AFSA based on particle swarm algorithm(PSOA) had been studied, which provided a reference for AFSA coalesced with other intelligent algorithms.Finally, the AFSA was applied to the robot path planning. The artificial fish individual was encoded by the way of reference points coordinate dimension reduction, which simplified the encode method of individual artificial fish and reduce the complexity of the algorithm. The path length and the collision probability with different security threshold were provided to policy makers, and the decision-makers could make decision according to the actual situation.