As you know very well the information sharing mechanism in PSO is significantly different from GA. They don't have genetic operators like crossover and mutation, particles update themselves with the internal velocity and they also have memory which is important to the algorithm, etc.
Could you please explain and introduce some papers which give full exposure to this subject and list the advantageous vs. disadvantageous, the similarity vs. differences of these two evolutionary techniques?
Hello Omidji, Her i have given few point on these two methods concentrating on advantages please go through this
The GAs strength is in the parallel nature of their search. The genetic operators used are central to the success of the search. All GAs requires some form of recombination, as this allows the creation of new solutions that have, by virtue of their parent’s success. In general, crossover is the principal genetic operator, whereas mutation is used less frequently. Crossover attempts to benefit offspring solutions and to eliminate undesirable components, while the random nature of mutation is probably to degrade a strong offspring solution than to improve it. The algorithms power is the implicit parallelism inherent in the evolutionary metaphor. By restricting the reproduction of weak off springs, GAs eliminates not only that solution but also all of its descendants. This makes the algorithm converge towards high quality solutions within a few generations.
Particle Swarm optimization shares many similarities with Evolutionary Computation (EC) techniques in general and GAs in particular. All these techniques begin with a group of a randomly generated population and utilize a fitness value to evaluate the population. They all update the population and search for the optimum with random techniques. The main difference between the PSO approach compared to EC and GA is that PSO does not have genetic operators such as crossover and mutation. Particles update themselves with the internal velocity; they also have a memory important to the algorithm. Also, in PSO only the ‘best’ particle gives out the information to others. It is a one-way information sharing mechanism, the evolution only looks for the best solution. Compared to GAs, the advantages of PSO are that PSO is easy to implement and there are few parameters to adjust.
Mahmoud,
I still find that the paper by Eberhart and Shi (Comparison between Genetic Algorithms and Particle Swarm Optimization) while very dated, it still offers the best comparison between the algorithms that I know of. Most recent papers that I have only focus on performance comparison but do not go into the details of what are the important differences.
Hope this helps
Please see this paper from MIT for comparison of GA and PSO.
Rania Hassan, Babak Cohanim Olivier de Weck - A COPMARISON OF PARTICLE SWARM OPTIMIZATION AND THE GENETIC ALGORITHM
Dear @Arturo and @Nageswara. Thank you for introducing the papers. Both are interesting.
According to the investigation from MIT group (2005), PSO outperforms the GA w.r.t. computational efficiency
An important distinction is that GA is initially a discrete technique that is also suitable for combinatorial problems, and PSO is a continuous technique that is very poorly suited to combinatorial problems.
To equate their search space explorations, the particle movement in PSO can be seen as a form of path re-linking among pbest positions. In this sense, both PSO and GA can be seen as generating new solutions in the neighbourhood of two parents -- via crossover in GA and via attractions to two pbest positions in PSO. This multi-parent effect should be a key difference (advantage?) over single-point techniques such as simulated annealing, tabu search, and 1+lambda-ES.
To discuss an advantage of PSO over GA, PSO is really two populations -- pbests and current positions. This allows greater diversity and exploration over a single population (which with elitism would only be a population of pbests). Also, the momentum effects on particle movement can allow faster convergence (e.g. when a particle is moving in the direction of a gradient) and more variety/diversity in search trajectories.
PSO is a parallell optimization strategy
GA a serial stratrgy
GA may be integrated in PSO
The most significant difference between GA and PSO regards to the definition of genes and particles. PSO comes from an agent oriented paradigm. Particles are semi-autonomous agent which see each other the status of each other and decide to change their status toward the best observed particle in their locality. But genes are not agent-like and don't have any abilities to sense their surrounding environment. In this case, there is an daemon program which selects and combines the genes toward the next generation of genes(solutions). The difference is in autonomy!
Thank you for this question. I agree that the connections between the two optimization techniques are very fragile. PSO simulates the "social behavior" of living communities (an interesting definition is given in VIKIPEDIA). GAs simulates the Theory of Darwin (after Holland). However, probably Genetics is still nowadays frequently more reliable than the social behavior.
Antonio Manuel, I have been used both GA and PSO for long time and I wouldn't say that GA is more reliable than PSO. However, I would like to understand better your point of view. Could you develop it a little further?
Dear Omar Andres. Congratulations! Are you following the world cup in Brazil (inside and outside of the pitch)? Unfortunately it seems that you did not fully understand what I meant. The basic ideas that are underlying to the original Genetic Algorithms technique sound. They are not based on empiricism (but on a scientific approach) and if duly modeled and implemented, GAs really work (the concepts of building blocks and parallel processing are really very powerful). However, I used the words “probably” and “frequently”. It means that it is not deterministically and that it may be not always.
I'm following up the world cup at distance because I'm in Canada up to 2015. But I do watch the games on TV. Brasil is doing well but I'm not sure if they have a team to get into the finals. The victory against Chile was a matter of luck and bad penalty shoots from Chile.
Speaking of GA and PSO, even having used those words I do not see reasons to say that GA is more reliable than PSO. I agree that GA has been more studied than PSO, but this happens because GA has more than 40 years and PSO was proposed only in 1995, which become a comparison in terms of studies quite unfair.
You can see to this publication where you find a comparative study of such meta-heuristics especially the PSO vs GA.
Conference Paper Application of the PSO Technique to the Optimization of CMOS...
I believe that PSO needs to be exercised more to improve its efficiency. Generally it is very slow process compared to conventional GA. And capability of GA is no lesser than any other optimization algorithms. It is very unique, very simple and always satisfying for large combinatorial problems. However, for less complex and continuous search problems PSO may perform better.
Dear Malay Thank you for your contribution. Regarding your saying "PSO is very slow process compared to conventional GA" I am not agree with. It can converge faster than GA. However, the points you made about GA simplicity, etc. are good points about GA.
Dear Mahmoud Omid:
Everything is relative and dependent of the type and of the dimension of the problem. For example (for the dimension N), an exponential time algorithm (a NP complete) is obviously worse than a polynomial time algorithm (we all have in mind multiple examples of huge numbers with exponential growing: neutrons bomb, bacteria, “lottery” alternatives, and so on). However, I learned a long time ago (particularly with the book of Alan Gibbons from 1985 entitled “Algorithmic Graph Theory”) that for a low dimension system, the first can perform better (as the result of the equation c^N can be actually lower than bN^c, where b, c and N are integer positives, while N is kept relatively small).
Also, frequently each problem requires the adaptation of the optimization techniques, including its association, for keeping or attaining the best performance. In the three-dimensional optimization of urban drainage systems (for example), the following paper illustrates well these fundamental ideas with an improved Dynamic Programming model connected to a modified Genetic Algorithm (associated with Graph Theory):
“Three-Dimensional Optimization of Urban Drainage Systems” by A. Freire Diogo, Godfrey A. Walters, E. Ribeiro de Sousa, Victor M. Graveto, Computer-Aided Civil and Infrastructure Engineering 12/2000; 15(6):409 – 426.
Best regards
António Manuel Abreu Freire Diogo
It depends on your problem. The logics used in Pso and Ga are different however PSO is faster than Ga but in traditional version both them have equal performance. Note that the original Version PSO was unable to solve discontinue prbls. Thus, it was hybridized in literature to have a better performance..
Thank you @Javad for commenting on this question. There are lots of comparisons among EA techniques in the literature. For example, the following paper by Elbeltagi et al. (2005) compares the formulation and results of five recent evolutionary-based algorithms: genetic algorithms, memetic algorithms, particle swarm, ant-colony systems, and shuffled frog leaping. Two benchmark comparisons (F8 [Griewank’s] and the F10 function) among the algorithms are presented for both continuous and discrete optimization problems, in terms of processing time, convergence speed, and quality of the results. The authors concluded that, the PSO was generally found to perform better than other algorithms in terms of success rate and solution quality, while being second best (after ACO) in terms of processing time.
Emad Elbeltagia, Tarek Hegazy and Donald Grierson (2005). Comparison among five evolutionary-based optimization algorithms. Advanced Engineering Informatics 19: 43–53.
If anyone has compared GA and PSO in case of optimum feature selection in Remote Sensing (SAR images), for the purpose of classification, could you please share your experience.
To improve the understanding of the scenario, I have a lot of features obtained from PolSAR decomposition and some are also from optical data (Landsat8). My target is to select the optimum features set from these two different data set which will bear complementary informations such that I can then use these feature subsets for classification (infact to identify a particular type of forest species). In this scenario which of the technique is more suitable and better (PSO, ACO, GA,...) and could you give me the promising reason why so?
As per my observation, PSO has the following advantages over GA: Simple concept, easily programmable, faster in convergence and mostly provides better solution.
PSO and GA are based on the same principle. A random element and the cost of error. They are useful for different applications.
I have used both approaches for the deployment of wireless sensor networks (which is an NP-complete problem). My conclusion: although PSO and the GA on average yield the same effectiveness (solution quality), PSO is more computationally efficient than GA.
Thank you for up to date answers. In fact, in many of recent investigations [e.g., 1, 2] I often read authors concluding that "PSO outperformed GA" when applied to various problems and PSO yields better quality and faster performance relative to GA. Although PSO and GA share many similarities, but PSO has no evolution operators (as crossover and mutation in GA) and potential solutions fly through the problem space by following the current optimum particles.
[1] Saed, A.A.A. and Kadir, W.M.N.W. (2011). Applying particle swarm optimization to software performance prediction an introduction to the approach. 5th Malaysian Conference in Software Engineering (MySEC), 2011.
[2] M. B. Abdelhalim and S. E. –D. Habib (2009). Particle Swarm Optimization for HW/SW Partitioning, in Particle Swarm Optimization. Edited by Aleksandar Lazinica. ISBN 978-953-7619-48-0
One of the main advantages of PSO algorithm as compared to other evolutionary ones is its easiness in implementation since there are no crossover, decoding and encoding of a genetic algorithmin a PSO.The computational time of PSO is inexpensive as compared to other optimal algorithms due to its limited number of parameters. PSO is simple both
in theory and numerically implementation. Because of its robustness, stability and simplicity.
Visit link for Matlab Code
https://youtu.be/OQ3T575sbI8
http://dg-algorithm.blogspot.com/
Brief about PSO over GA
Ø Particles update themselves with the internal velocity; they also have a memory important to the algorithm. Also, in PSO only the ‘best’ particle gives out the information to others. It is a one-way information sharing mechanism, the evolution only looks for the best solution. Compared to GAs, the advantages of PSO are that PSO is easy to implement and there are few parameters to adjust.
PSO Adv
Ø Simple concept, Easy implementation
Ø Robustness to control parameters
Ø Computational efficiency
Ø Insensitive to scaling of design variables
Ø Simple implementation
Ø Easily parallelized for concurrent processing
Ø Derivative free
Ø Very few algorithm parameters
Ø Very efficient global search algorithm
In many research papers, it is proved that PSO could provide faster convergence and could find better solutions when compared to GA. The implementation of PSO is also simple. But the main disadvantage of PSO is with its poor local search ability. To improve the diversity in the local search space PSO can also be combined with GA as a hybrid algorithm.
The execution time of the GA is higher than PSO, and the convergence is slower. PSO converges fast because it can be implemented without too many parameters and operators.
GA has high Computation efforts compare to PSO. However PSO has tendency of taping in local best. GA has better Std. Deviation for fitness than compare to PSO.
In general terms, PSO tends to have premature convergence according to its search strategy. On the other hand, GA maintains a better performance, since its operators allow to avoid get trapped in local minima.
(+)PSO: Easy to understand, easy to program, and convergate fast
(-) PSO: Fall tend to fall into a sub- optimal state
(+)GA: Evolutionary programing problems, parameter optimalization
(-)GA: Computing power
In its original version, it seems worst than GA. However, new versions avoid the premature convergence and intersitin capacities of exploration and exploitation.
https://www.researchgate.net/post/Using_of_PSO_Particle_Swarm_Optimization_and_GA_Genetic_Algorithm_depending_of_the_optimization_problems_what_is_suitable_algorithm_to_use2
https://www.sciencedirect.com/science/article/pii/S0377042710004796
https://pdfs.semanticscholar.org/c6d7/300bac7894c5b77d7abcab08a14c4632775c.pdf
•Simple concept, Easy implementation
•Robustness to control parameters
•Computational efficiency
•Insensitive to scaling of design variables
•Simple implementation
•Easily parallelized for concurrent processing
•Derivative free
•Very few algorithm paras
•Very efficient global search algorithm
Please refer to this paper
Article A review of population-based meta-heuristic algorithm
there are three parameter to comparison this two methods:
1. reliability: average best score
2.accuracy: 20% average best score
3.run time: total computation time
GA has less run time and more reliability and accuracy for discreet parameter.
PSO get stake in to a local minimum and has more run time.
Please refer to these papers:
https://www.irjet.net/archives/V3/i6/IRJET-V3I622.pdf
Also, you can find a deep analysis of PSO in the following paper before section 4.
Article CCSA: Conscious Neighborhood-based Crow Search Algorithm for...
I think PSO is more improved related to GA. The computational cost of PSO is lower than GA. It means that the number of function evaluation of PSO is much lower than GA. Furthermore, in PSO method, fewer points fall in to local area and more reliable answers are obtained.
Pso is faster than GA in terms of convergence. But GA is better in avoiding local optima value. In general it depends on how you have tuned your parameters.
When two nature-inspired methods are compared, we can never say that method A is better than method B. method A may be good at exploration but bad at exploitation. So there is some trade-off every time and that's why there are so many algorithms to tackle different aspects of the optimization problem. Similarly, PSO has expertise in some areas which are lacking in GA. For example:
1) PSO has an inbuilt guidance strategy which lets the solutions in PSO to obtain useful information from the better solutions and thereby helping them improve their own solutions. This results in faster convergence for the solutions in PSO. GA has no such guidance mechanism. Better solutions only pass the information when they participate in crossover with some other chromosomes.
2) PSO uses memory to store the previous best solutions obtained by every candidate. This helps the candidates to recover their solutions when they get diverted to an unwanted direction. So, when the candidates start exploring an unwanted path and the quality of solutions starts degrading they can revert and get directed to the previous stage through the pBest (personal best) component in the velocity term. This makes the algorithm very robust and susceptible to degradation. On the other hand, GA does not use memory to keep track of the solutions throughout different generations.
3) Modern machines are really good with mathematical computations. PSO has an inherent tendency to exploit this advantage because all the computations and procedures in PSO are purely mathematical. But, GA has procedures like crossover, mutation which are not purely mathematical which makes GA more time-consuming than PSO.
Hope this helps. Best of luck.
PSO has an intrinsic guidance approach which lets the solutions in PSO to get useful data from the improved solutions and thus helping them develop their individual solutions. It gives a faster convergence rate for the solutions in PSO. But GA is not having such a guidance mechanism.
Meta-heuristic techniques such as PSO and Genetic Algorithms are inspired by nature and have proved themselves to be effective solutions to optimization problems. However, their performance is depending on parameter tuning. Jones (2005) wrote an article in which PSO and GA are compared from different points of view. For better understanding, you can review their work. To my experience, although both algorithms can be employed in both continuous and discrete problems, they illustrate different outcomes. The PSO algorithm outweighs GA in the continuous problem while GA is superior to PSO in the discrete optimization problems.
Particle group optimization (PSO) is a relatively new innovative search method Its mechanics are encouraged by the group or collective behavior of biological communities. PSO operates similarly to genetic algorithm (GA) in that two innovative evolutionary methods are considered population-based search methods. In other words, PSO and GA go from one point (population) to another in a iteration and are promoted using definite and probabilistic rules. And many versions of it have gained a reputation in academia and industry for their understanding, ease of execution, and ability to effectively solve properly integrated, nonlinear optimization problems that exemplify complex engineering systems. GA forms have a very high computational cost
PSO is faster, more reliable and more effective in handling continuous problems.
The execution time of the GA is higher than Particle Swarm Optimization (PSO), and the convergence is slower. PSO converges fast because it can be implemented without too many parameters and operators.
PSO is a fast gradient method. GA is a evolution method.
The methods have different applications
PSO has faster convergence than GA, GA takes a lot of time to simulate the generations function value.
Unlike GA, the variables in PSO can take any values based on their current position inthe particle space and the corresponding velocity vector.Genetic algorithms do not handle complexity in an efficient way, because in suchcases the number of elements undergoing mutation in very large which causes aconsiderable increase in the search space. So, in this case PSO is the best alternativeas it requires small number of parameters and correspondingly lower number ofiterations.GA usually converges towards a local optimum or even arbitrary points rather thanthe global optimum of the problem while as PSO tries to find the global optima.
- Badges
- Science topic
- Similar topics
- Computer Communications (Networks)
- Simulators