Like other meta-heuristic algorithms, some algorithms tend to be trapped in low diversity, local optima and unbalanced exploitation ability.
1- Enhance its exploratory and exploitative performance.
2- Overcome premature convergence (increase the fast convergence) and ease of falling (trapped) into a local optimum.
3- Increase the diversity of population and alleviate the prematurity convergence problem
4- The algorithm suffers from an immature balance between exploitation and exploration.
5- Maintain the diversity of solutions during the search, so that the tendency of stagnation towards the sub-optimal solutions can be avoided and the convergence rate can be boosted to obtain more accurate optimal solutions.
6- Slow convergence speed, inability to jump out of local optima and fixed step length.
7- Improve its population diversity in the search space.
Refer to the papers by Dr. GaiGe Wang, Dr. Amir Alavi and Dr. Amir Gandomi. They are renowned experts in Metaheuristic Optimization.
Dear Essam H. Houssein
These papers might be helpful, have a look:
1. Article A Review on Genetic Algorithm: Past, Present, and Future
2. https://res.mdpi.com/d_attachment/applsci/applsci-10-07000/article_deploy/applsci-10-07000.pdf
3. https://www.degruyter.com/document/doi/10.1515/jisys-2020-0117/html
Kind Regards
Qamar Ul Islam
Understand you mean "lack of diversity". Not sure if this is a genetic/evolution type algorithm or data driven. If you have a population where each individual is represented somehow, I would store the data at succesive (time or step) check points (i.e. dataset, one record for each population member with key characteristics). Then you can analyze this and see the population distribution for different characteristics and at different times during the evolution. Once you know that, you can maybe introduce some functionality in the code to influence the population evolution and diversity, etc. ..
Initially the population is usually random to have diversity. Later on it is required to exploit towards some local optimum. Also add some diversity additionally to avoid pre-mature convergence.
That is a tricky question!,
I think that there is not such as a stright answer, instead you should look to the problem because each metaheuristic variates its behaviour regarding to each problem "Problem-depdendent", I know that there are methods that are characterized by its exploitative behaviour but at the end everything is related to the problem (and the configuration of the algorithm).
I find your question tricky because if someone says that an algorithm "X" suffers of premature convergence then it means that such algorithm was tested in all the universe of problems which is not possible, in fact there is present the "No free lunch theorem (NFL)". Note that the terms exploration and exploitation are not defined in a standard way in literature and there are few papers that properly defines such words (right now come to my mind the Matej Crepinsek and Sthepen Chen works).
However, if I wanted to classify populational meta-heuristics I would try to test several problems with different properties agains several algorithms (with several runs of course), then report several metrics of diversity against time, in literature there are a lot papers that explain metrics that are devoted to measure the diversity of a population, in fact it might be applied any sparsity metric (e.g. I often apply the ADI-Average Distance between Individuals).
I hope this helps,
Bests!
A possible strategy to deal with this problem is the adaptation of a Boltzmann selection, which has the guarantee of selecting better energy states (fitness function) than the previous population and avoiding a convergence to the local optimum (de Anda et al. 2019)
Dear Dr. Essam H. Houssein ,
One of the ways is to check the performance of specific metaheuristic search strategy on standard test functions. As the global Pareto fronts of these functions is known, it can be compared with non-dominated front derived by the metaheuristic search strategy, using multiobjective performance indicators. In this way many of your concerns can be addressed.
Best,
Dr. Abhinav
https://www.mathworks.com/matlabcentral/fileexchange/92150-sperm-swarm-optimization-sso#:~:text=A%20new%20meta%2Dheuristic%20optimization,low%2Dtemperature%20zone%20called%20Cervix.
https://github.com/sh7adeh1990/MOSFP
https://au.mathworks.com/matlabcentral/fileexchange/92130-hssogsa
I think the test functions such as sphere, Rosenbrock, Griewank,
Rastrigin, NC-Rastrigin and Ackley functions best choice to check the performance of your algorithm.
https://www.sciencedirect.com/topics/computer-science/benchmark-function
like Mr . Joel Chacón, I think also that question is a dependent problem, there is no exact answer.
By benchmark functions, especially multi-modal functions such as Rastrigin.
You can use the following benchmark functions to assess the performance of your algorithm
https://www.mathworks.com/matlabcentral/fileexchange/106450-new-hard-benchmark-functions-for-global-optimization?s_tid=srchtitle
You can consider the findings reported in this article and study the diversity curves generated for your problem.
https://www.sciencedirect.com/science/article/abs/pii/S2210650219304080
Some remarks in the following paper may reply your question:
Article Design, Implementation, and Optimization of Sliding Mode Con...
- Badges
- Science topic