Research On Evolutionary Machine Learning Algorithm And Its Application In Earthquake Prediction

Evolutionary algorithms are a series of global optimization algorithms inspired by biological population behavior.Compared with traditional algorithms such as gradient optimization,evolutionary algorithm is simple to implement,has higher flexibility and robustness,so it has become an indispensable part of modern optimization algorithms.In recent years,many novel evolutionary algorithms have been proposed,including moth-flame optimization(MFO),gradient-based optimizer(GBO)and heap-based optimizer(HBO).These algorithms make use of the advantages of swarm intelligence to get satisfactory results when dealing with complex problems.However,with the expansion of the application scope of evolutionary algorithms and the continuous crossintegration with other disciplines,these algorithms are prone to fall into local optimization and slow convergence speed when dealing with practical problems.In this paper,MFO,GBO and HBO are improved,and the performance of the improved algorithm is analyzed in many aspects.The improved algorithm is also applied in the field of earthquake prediction.Research contents include:(1)Based on MFO,an improved weight moth-flame optimization(WEMFO)algorithm is proposed.WEMFO adaptively adjusts the search strategy at different stages of the algorithm,making the algorithm more flexible and reliable in the transformation between global search and local search.WEMFO is compared with other evolutionary algorithms in recent years on 30 benchmark functions,and the expansibility of WEMFO algorithm is analyzed.(2)On the basis of GBO,gaussian bare-bones mechanism(GBB),oppositionbased learning(OBL)and moth spiral updating mechanism are introduced.An improved gaussian opposite moth-flame gradient-based optimizer(GOMGBO)is proposed.In GBB,the idea of differential evolution algorithm is applied,and the branch selection structure is adopted at the same time of gaussian variation,so that the vector after gaussian variation is not limited to the local optimal solution.The essence of OBL is to deal with the opposite side of things at the same time.The spiral updating mechanism of moths simulates the night flight pattern of moths and is the core of MFO,which ensures GOMGBO to find the global optimal solution without falling into the local optimum.GOMGBO is compared with many famous basic evolutionary algorithms and improved evolutionary algorithms on 30 benchmark functions.(3)On the basis of HBO,modified rosenbrock's rotational direction method(MRM),grey wolf mechanism(GWM)and orthogonal learning(OL)are introduced,and a new improved HBO algorithm(gray-wolf orthogonal heap-based optimizer,MGOHBO)is proposed.MGOHBO is compared with 11 well-known primitive and improved meta-heuristic algorithms on 30 benchmark functions.(4)The improved evolutionary algorithm is applied to earthquake prediction.WEMFO,GOMGBO and MGOHBO algorithms are used to optimize two parameters of kernel based extreme learning machine(KELM)model respectively,and then average the parameters obtained by the three algorithms.Forming a new hybrid KELM model for earthquake prediction.