Research On Adaptive Evolutionary Algorithm For Complex Problems

With the rapid development of computer technology,optimization problems involve all fields of human learning and life,and become more and more complex and changeable with the progress of society.Most optimization problems can be solved through abstraction and modeling,and how to effectively solve these complex problems has become a challenging research topic.Evolutionary algorithm,as a branch of artificial intelligence involved in the optimization field,with its advantages of self-adaptation,self-learning and self-organization,can not be limited by the nature of the problem,and effectively solve most of the highly nonlinear and multi-peak optimization problems in the real world.However,"No Free Lunch" theorem from the macro level concern optimization algorithm in the complex set of average performance,further pointed out that the performance of the algorithm to solve the problem of related,namely the algorithm of a comprehensive performance is very poor,if on the test problem of a matching degree is high,still can show more than any other performance of the algorithm,and also the opposite.Thus it can be seen that the test problem is the standard to evaluate the performance of the optimization algorithm,while the optimization algorithm must rely on the target optimization problem to present its effect and performance.The two complement each other and become a unified whole.However,there are some problems in the optimization algorithm design and problem design in the field at present.On the one hand,the evolutionary algorithm is not designed for the fixed characteristics of the problem,which is not intelligent and flexible enough to effectively solve the properties of complex problems.Therefore,it is necessary for the algorithm to solve different problems adaptively.On the other hand,although the existing test problems use a variety of transformation technologies to increase the difficulty of optimization,they are less complex than the real world problems,and will not bring too big obstacles to optimization to a certain extent,so as to better detect the overall performance of the algorithm.In order to solve the above problems,this paper,inspired by intelligent computing and graphic transformation,integrates the ideas of neural network,reinforcement learning,local and global search,and explores and studies the evolutionary algorithm and test problem design.The main research contents of this paper mainly include the following aspects.(1)A new conformal mapping method based on benchmark function is proposed.By bending the shape of the benchmark function and increasing the complexity of the problem,the performance of the optimization algorithm can be evaluated better and more comprehensively.In addition,the influence of the parameters involved in the conformal mapping process is also explored,and the performance of each parameter on the shape,size and optimal position of the ring is analyzed.(2)Aiming at the complex pathological problems,a novel particle swarm optimization(PSO)algorithm with reinforced velocity search strategy is proposed,through directly learning and correcting the optimal velocity to find the optimal value.At the same time,it also refers to a lot of information to help algorithm be more valuable mobile and explore more valuable,increasing the probability of particles to find better direction and improve the exploring ability of algorithm,so as to find the optimal solution more efficiently.(3)In order to improve the adaptive ability of the algorithm on complex and changeable problems,an adaptive evolutionary algorithm based on neural networks is proposed,which makes better use of the strong fitting ability and adaptive ability of neural network as well as the cooperative and competitive relationship between populations,and adaptively carries on the next update through state and action feedback.Experimental results preliminarily show that this method has learned the ability to solve problems,and can adaptively transform an unknown optimization problem into a solvable problem when faced with it,showing a strong potential to capture the characteristics of test problems.