The reason why the high-dimensional complex single-objective function optimization problem is difficult:
(1) With the increase of the dimension, the search space increases exponentially. Even if the population size of the algorithm is determined, the proportion of high-quality feasible solutions is relatively small, so that the probability of obtaining a direct reduction is reduced. It is understood in this way. The purpose of increasing the overall sense dimension is to obtain better solutions, that is to say, as the latitude increases, the solution is more constrained, and the probability of obtaining a high-quality solution is smaller)
(2) The optimization time of the swarm intelligence algorithm is relatively long, that is, the optimization accuracy will be affected by the number of evolutions. Under the constraints of time, even if the algorithm can cover everything, the realistic results may prompt giving up. Label it with low practicality (like 3 seconds in web pages, usually the maximum threshold for customers to wait)
(3) As the dimension increases, the convergence speed of the algorithm will decrease, so that the algorithm can no longer complete the operation in a limited time.
(4) The existence of multiple local optimal solutions
Logical Mode:
* ***Algorithms are also unavoidable of the above problems . Through experiments and analysis, it is found that there are two reasons why the algorithm is prone to fall into local optimum : First, the search behavior of the entire wolf group is overly dependent on the position of the head wolf, which is easy to cause prey. The diversity of the group decreases; second, the wolf group algorithm has many parameter settings, which are not easy to control. If the parameter settings are not appropriate, it is easy to cause the algorithm to have low optimization accuracy or non-convergence. There are two reasons for the slow convergence of the algorithm: first, the optimization process of the algorithm is divided into wandering, raiding and siege, but the lack of necessary information exchange between artificial wolves makes the algorithm not global and too scattered; Second, the step size of the siege behavior is constant. If the step size is not set properly, the algorithm will not be able to accurately siege the target, resulting in a decrease in the convergence of the algorithm. For the above reasons ................. + Design thinking
1. Interactive wandering behavior and interactive summoning behavior are designed, so that wolves and fierce wolves can exchange information, which helps artificial wolves to better grasp the overall information, ensure the diversity of wolves, and improve the ability of wolves to search for optimization; in order to speed up Convergence speed, an adaptive siege behavior is designed for the siege behavior of wolves. According to the distance of the wolves from the prey, the siege step length is adaptively adjusted, so that the algorithm has strong adjustment ability.
The main improved strategies (1) interactive strategy of roaming behavior (2) interactive strategy of rushing behavior (3) adaptive strategy of wolf siege
interactive walk
(1) The wolf detection is randomly selected in h directions and then carried out, but there is no communication with each other. Is there a repetition between them? You can record the direction and location you have found. Use the gbest guide role in the article. formula:
In the initial walk, it is to obtain prey, so in the process of swimming, according to the concentration of smell, you will get to see the prey, and then according to the prey
Fake code:
Interactive raiding behavior
Adaptive Siege Behavior
Overall dissertation in ch
When I saw the improvement of the paper, I felt it was very reasonable. Although I have finished reading the later ones, but I am not in the state, and some cannot understand. When interactively roaming, that gain is the interactive prey? as well as spawning h+1 prey. Don't know yet