The PSO algorithm will require the calculation of objective functions as it proceeeds. The ANN network can be developed with training and testing data sets using the error estimates. The trained network can be supplied as objective function to the PSO:
F = sim(net, X') OR similar expressions.
Actually, I don't have any document. I use the Radial basis neural networks (RBNN) and genetic algorithm for optimization.
I your case, you use the PSO, but it will require function evaluations for which you need to formulate the a function of the form Y = F(x). Since, I don't know how you obtain the ANN network, it is difficult to obtain an exact expression.
In the second case, if you want to train the ANN using PSO, it is another application of PSO. I attach a file for your reference.
Dear Kazeem Alli,
Please read the paper of Geeraerd et al., as it provides great insight not only about the joint use of PSO and ANN, but also about the error analysis of the free optimization parameters:
Geeraerd, A.H., Herremans, C.H., Ludikhuyze, L.R., Hendrickx, M.E., Van Impe, J.F., «Evaluation of Model Parameter Accuracy by Using Joint Confidence Regions: Applications to Low Complexity Neural Networks to Describe Enzyme Inactivation», Mathematics and Computers in Simulation 48 (1998) 53-64.
http://pbil.univ-lyon1.fr/members/lobry/articles/AEM_1995_61_610.kim/GeeraerdAH1998.pdf
For more information about the calculation of joint confidence regions with PSO (or other evolutionary algorithms) read the paper of Schwaab et al.:
Schwaab, M., Biscaia Jr, E.C., Monteiro, J.L., Pinto, J.C., «Nonlinear Parameter Estimation through Particle Swarm Optimization», Chemical Engineering Science 63 (2008) 1542-1552.
http://www.sciencedirect.com/science/article/pii/S0009250907008755
Good luck.
Dear Kazeem Alli,
Each Artificial Neural Network (ANN) has a numbers of parameters in forms of weights and biases which are supposed to be effectively optimized in order to response accurately into an input record which can include some numbers representing different affecting parameters on the target.
There are different conventional and modern methods which can be gained to optimize the referred parameters such as Delta rule, Genetic Algorithm and Ant Colony. By the way, nowadays it can be observed that trends of applying modern population based algorithm like Bee Colony and PSO are popularly followed by researchers.
In fact, the PSO and ANN are coupled together so that the PSO acts as a supervisor to optimize the ANN relevant parameter. I and my colleagues have done some papers about application of the mentioned techniques in petroleum engineering which could found in my profile . Then, the optimized forms of PSO-ANN could be applied to predict the supposed parameter.
To answer you question, you need firstly to gather similar data from other likewise reservoir from all over the world, and after that based on the gathered data set you must develop the best structure of the ANN which in this case you can gain from previous literature. Next, it is the turn to optimize the weights and other parameters by gaining from PSO. After finishing all above steps, apply the concluded form of PSO-ANN on some parts of your database known as testing part. If the generated results have a good performance, it is determined by using statistical indices, it has the capability of being applied in your case, for example making prediction of ICV.
If you have any probable question, do not make any hesitation to ask me.
By regards
@Mohammad, thanks for the answer but i need documents that can make me get the full understanding in terms of practicability and your detail contact.
https://www.researchgate.net/publication/297245624_Particle_Swarm_Optimization_Algorithm_and_its_Codes_in_MATLAB
Research Particle Swarm Optimization: Algorithm and its Codes in MATLAB
https://www.researchgate.net/publication/296636431_Codes_in_MATLAB_for_Particle_Swarm_Optimization
Code Codes in MATLAB for Particle Swarm Optimization
- Badges
- Science topic
- Similar topics
- Engineering
- Petroleum Engineering