My research problem is about resource searching in mobile ad-hoc networks. I would like to design a soft computing based resource discovery model for this problem.
I want to draft a GA-like method but am stuck further.
GA way algorithm:
1. randomly guess 100 x and y values (population size is 100)
2. for each generations
3. find fitness y=f(x,y) for each pair of x and y. - you will have 100 y values.
4. take two x,y pairs for which you got highest y values (this is the fitness function here - highest amplitude is the fitness)
5. So, the selected two location may be near to a local maxima or global maxima.
6. using that two x,y pairs, generate another 99 x,y pairs (by keeping one previous best solution)
7. find fitness y=f(x,y) for each pair of x and y.
8. if the maximum y of previous generation is almost equal to present generation, then terminate here else go to step 4 for N generations.
9. At the end of the above N generations or at the terminating condition at 8, the GA will possibly find the global maxima (by randomly jumping here and there and linearly traversing the surface with logic)
(I am stuck up here as i am unable to fit any logical way of finding this y using some x in a GA like manner).
(Further i am unable to express a fitness function with the following design variables).
Note:- At each node, we will only have two information i.e., the neighbor list and neighbor count(number of neighbors). Further at each node, we can also estimate its local parameters such as mobility and remaining battery energy etc., - but a node do not know the mobility and remaining battery energy of other nodes.
Firstly, i would like to know whether it is feasible & correct to apply GA here -- if the search landscape change within the time-range of the search (i.e., after you started the heuristic but before you decide to stop and consider its result)
If i can apply GA kindly help me to fit a fitness function here.
Kindly give me some suggestions and feedback.
Thanking you in advance.
Your's truly,
ajay
Dear Ajay,
I am not entirely clear about the structure of your problem but if you are unable to express a fitness function based on your design variables then you cannot run an optimisation. In other words, if you only have access to nodal information that cannot help you decide which action you should take then an optimiser cannot help you either.
Let's try this: With the information you have about your problem, can you guess a good solution without running an optimiser? If yes, try to mathematically express how you did it in order to obtain a fitness function suitable for your problem.
Sorry I couldn't be more helpful. Good luck.
Terence
Genetic algorithm (GA) maintains a population of individuals, for generation. Each individual represents a potential solution to the problem.
Each individual evaluated to give some measure of its fitness. Some individuals undergo random transformations by means of genetic operations to form new individuals.
There are two type of transformation :-
1. Mutation, which forms new individuals by making changes in a single individual.
2. Crossover, which forms new individuals by combining parts from two individuals.
Your problem statement is not clear. I do not understand what the (x,y) pair is supposed to represent (a node in the network?) and I do not understand what you mean by "fitness" of the node. Fit for what, exactly? What "resource" are we trying to find? What makes one (x,y) location better than another one?
A genetic algorithm is not good for finding a needle in a haystack; it needs some sort of relatively smooth and mostly connected landscape, such that it can start at a random place and then feel its way in a random fashion to take steps uphill to a peak (or in an inverted landscape downhill to a lowest point).
You need to describe your landscape before you start devising your algorithm. Figuring out how to make your landscape suitable for a GA will tell you what your fitness algorithm is: In fact, every GA should begin there, with a landscape that gives you a fitness metric as a single floating point number, and then you can figure out the specifics of the mating and generational dynamics.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Algorithms