I have not done that specific of work, but seems like a classical prediction or maybe a kind of classification work.

Firstly, i would assume that, when you are talking about "permeability" you are talking about physical characteristic of a mean (material 1) that let to cross through it self something more (material 2). And that this last happen because material 1 has pores big enough like to let molecules of material 2 pass.

If I remember well, permeability (in the most general sence) depends on several variables like porosity, intergranular surface area regarding material 1. Density, volume about material 2 and time and preassure like independent variables.

If you want to attack this problem using ANN as prediction problem, in genera, you have to have the set of data, this means, a data base with experimental data that should cointain (as I can see just right now):

1)Characteristics of material 1: Features that define material.

2) Characteriscts of material 2

3) A meassure of material 2 permeaded through material 1.

Having this data, you should divide your set of data in 2 groups: a)training set (70%) and b) production set (30%).

Once divided your data set, you should to practice a pretreatment to the data to get them in a normal range.

With your data treated this way you can start to train your ANN using the training set, and finally to probe your ANN with the other 30%.

In theory, if your data set is "representative" your ANN would be able to recognize and predict permeability on new data that it never had "seen"

I knew recently this kind of work (permeability in oil) has been done using Genetic Algorithms.

regards

P.S Please forgive my ugly english :)

Witch permeability ?

With work specific to fuzzy logic and applications with genetic algorithms.

Limited experience predicting permeability using conformal predictors

Well... If this is a fuction and if you want to extrapolate .... Then you can try it ... If it is no functional behaviour ... You wont be able to predict ....

I have not done that specific of work, but seems like a classical prediction or maybe a kind of classification work.

Firstly, i would assume that, when you are talking about "permeability" you are talking about physical characteristic of a mean (material 1) that let to cross through it self something more (material 2). And that this last happen because material 1 has pores big enough like to let molecules of material 2 pass.

If I remember well, permeability (in the most general sence) depends on several variables like porosity, intergranular surface area regarding material 1. Density, volume about material 2 and time and preassure like independent variables.

If you want to attack this problem using ANN as prediction problem, in genera, you have to have the set of data, this means, a data base with experimental data that should cointain (as I can see just right now):

1)Characteristics of material 1: Features that define material.

2) Characteriscts of material 2

3) A meassure of material 2 permeaded through material 1.

Having this data, you should divide your set of data in 2 groups: a)training set (70%) and b) production set (30%).

Once divided your data set, you should to practice a pretreatment to the data to get them in a normal range.

With your data treated this way you can start to train your ANN using the training set, and finally to probe your ANN with the other 30%.

In theory, if your data set is "representative" your ANN would be able to recognize and predict permeability on new data that it never had "seen"

I knew recently this kind of work (permeability in oil) has been done using Genetic Algorithms.

regards

P.S Please forgive my ugly english :)

I do ANN, but not on permeability, I do prediction and clasiffication on other phenomenon.

Why do you ask? I mean, Do you want to compare results? or you want to interchange insight?

Hi Mahdi, can you explain better what you need? What kind of data you will need to process? If will need to predict something on software, you will need a database to search and compare some situations... for some specific problems, you will can use agents with case-based reasoning and fuzzy logic. I'm using JADE framework to develop agent systems and it works **very** well. ;-)

Thanks for asking, although I have no idea about the topic, but for concern I googled it out, which I am sure you must have also done it. I found something, may be useful:

1. http://www.onepetro.org/mslib/servlet/onepetropreview?id=SPE-99286-STU&soc=SPE

2. http://www.civilica.com/EnPaper-IOGPC17-IOGPC17_086.html

and 1 pdf attached

Please reply if I helped you

Neither have I - just wondring why you think neural network will be the best 'tool'?

Often when you ask a question that is too specific you get negative results. People try to be helpful, but there is a lot of research out there, and very little overlap between researchers. Please tell us more about your own research, and those of us with parallel experience can be more forthcoming about how our research is similar. Don't expect that everyone has the same exact background as you do.

Any neural network model can do a permeability prediction. The reason for that is that ANN or RBF NN are great predictors and do as well as or even better than any other predictive model in many domains.

i have no idea about permeability, but i used RNN in prediction. RNN( recurrent neural network) is better to predict thr value for longer prediction horizon, if u want to predict for small prediction horizon (5 to 45 minutes), u can then use feed foward neural network.

Training of neural network needs huge input-output data. This data can be collected by establishing input-output relationships by utilizing either DOE or Taguchi technique. You can refer my paper mentioned below. In our work permeability is one of the output in green sand mould system. Both BPNN and GA-NN algorithms have been used.

"Forward and reverse mappings in green sand mould system using neural networks"

AuthorsMahesh B Parappagoudar, DK Pratihar, GL Datta

.Publication date2008/1/31

.Journal nameApplied Soft Computing

.Volume8

.Issue1

.Pages239-260

.PublisherElsevier

.

Dear Faiz,

Alberto Hernández explained the general principal very well. Now this can be applied to your problem. If you can share your exact work/problem then maybe I can help you. I need to know where you are planning to apply the principal. Researcher mostly recommends it for Identifications security threats like who can pass though you security walls mean who is permeable. It has many other weighted parameters based on which one can decide the permeability of incoming security attack. But this model can be applied in numerous other situations. Kindly let me know you Ares where you are planning to apply.

Regards

Prof Santosh Bothe

This problem can be solved by the previous works are such that M-P model, PDP model, and Connectionist model, and others.

I think that the problem may focus on the activity of neurons in neural networks, for example the patterns of input-output in long-term memory, in short-term memory and working memory.

Thanks - seems to me that a rule based system combined with web search of chemical/biochemical/physical databases will solve the unknowns in a given combination faster than a neural network - just my aproach - Out for me

As ANN is a good predictor It can be applied for permeability prediction. The all you require is the input-output training data patterns.

Dear Faiz,

I have extensive and wide experience of the application of AI in petroleum reservoir characterization which includes the prediction of permeability. I have used more recent techniques and my present study is in the hybrid and ensembles of AI techniques.

May I know exactly what you need help on?

What exactly is your problem ?

As an addition, to the reply by Alberto Hernandez, do not forget the most important data set, in addition to the training and validation set : the test set. Once you have chosen the complexity of your machine (i.e. the number of hidden neurons if you use a neural network), you must test your machine on the test set, which contains data that is used neither for training nor for validation. This is the only statistically meaningful way of assessing the performance of a learning machine (neural network or other).

I have think seriously on the commet by Anders Nielsen in this discussion (interesting at all).

What he said made me think one important topic that its the quality and availability of data. This two last factors are decisive on the choosing of the paradigm or method to solve the problem.

Maybe, if you want to share some more details about the problem we could (together) help.

Quality and availability of data is a good beggining to start and evaluate the nature of your ploblem.

Any detail about?

Best regards to everyone.

I am working on a model of Mind:The Elementary Pragmatic Model (you can see on Google). I did a program, sill not fully completd that work on processes and not on contents. a lot of ideas come up from the program. These ideas are not the absolute truth but Just intersting ideas ( Then we are in the field of problem solving creativity). Then I compare these new ideas with ideas that are allready in my mind. From these comparison emerge new pathway of thinking, new way of follow goals creatively. I think this process iis very importantant.I use the Elementary Pragmatc Model in Psychotherapy ( De Giacomo et al. Rivista di Psichiatria 47,1,2012), to ameliorate the functioning of Firms and to Write articles for newspapers ( De Giacomo e coll in Luciano L'Abate Editor De Giacomo et al. "Paradigms in Theory Construction": .:Informtion Processing. Chapter 18, 341-363,2012. Springer

There is some background to this in metal detection technology. With metal detectors there are two flavours of professional detectors: VLF IB and PI. BFO detectors faded into history. VLF detectors run at frequencies ~10kHz and use separate coils for Tx and Rx that are balanced and thus orthogonal, and ground permeability (etc.) is tackled with something called "ground balance" by means of phase shift/delay. PI detectors are completely different kind of animal and so far are not friendly with "ground balance" at all. There is a patent granted to White's company that combines VLF and PI, where VLF is used mainly for ground balance, so there you have it.

You'll find metal detection technology burdened by worst possible and mostly ridiculously unreadable patents of all, but in case you wish to scratch a bit into that direction patents are your only choice.

Hi all thank you very much for all of you.

Here you are more information on my work of permeability prediction using ANN model for anyone who is interested to help more.

I am trying to predict solid particles permeability using liquid (water), this solid particles are less than 10 microns in size with different solid concentration, particles shape and size distribution. However, different experimental processes are including such as filtration, permeation and sedimentation that affect permeability very much and have almost 9 magnitudes different between filtration (lowest permeability) and sedimentation (heist permeability). i have collect around 550 sets of data for the four input parameters (size, size distribution, concentration, shape of particle) also the measured permeability from Lab. work. i did my ANN code and it is work very well with very low magnitude (5) but ones i use all magnitude rage (9 different) it does not predict a good results. it is error ratio is very high and regression is very low. i tried to work on pre-process methods such as normalization or PCA function. Normalization gave a good results but i could not work well with PCA function.

So, i am looking for someone to help with pre=processing methods and to check the data if we can do any more statistical work with it before we feed to the ANN model finally, to check my work in general data and code if i have done any mistakes with coding.

Regards

May be this can help

Estimations of porosity and permeability from well logs are important yet difficult tasks encountered in geophysical formation evaluation and reservoir engineering. Motivated by recent results of artificial neural network (ANN) modelling offshore eastern Canada, we have developed neural nets for converting well logs in the North Sea to porosity and permeability. We use two separate back-propagation ANNs (BP-ANNs) to model porosity and permeability. The porosity ANN is a simple three-layer network using sonic, density and resistivity logs for input. The permeability ANN is slightly more complex with four inputs (density, gamma ray, neutron porosity and sonic) and more neurons in the hidden layer to account for the increased complexity in the relationships. The networks, initially developed for basin-scale problems, perform sufficiently accurately to meet normal requirements in reservoir engineering when applied to Jurassic reservoirs in the Viking Graben area. The mean difference between the predicted porosity and helium porosity from core plugs is less than 0.01 fractional units. For the permeability network a mean difference of approximately 400 mD is mainly due to minor core-log depth mismatch in the heterogeneous parts of the reservoir and lack of adequate overburden corrections to the core permeability. A major advantage is that no a priori knowledge of the rock material and pore fluids is required. Real-time conversion based on measurements while drilling (MWD) is thus an obvious application.

thanks to all

I have done some correlation work to find the most effecting parameters and I got these four (size, size distribution, shape, concentration of particles). But I still have problem of using PCA function with my data as a pre-process step before feeding data to the ANN model. Could you help on this? I am a chemical Engineering researcher I am not professional on MATLAB or ANN. I have done some work with both but I still have difficulties on my ANN code. Could you tell me how can you help me on this work?

Regards

Faiz,

Simple MLPs work quite well for molecular permeabilities through biological membranes:

http://www.simulations-plus.com/Definitions.aspx?pID=13&lID=56

http://www.simulations-plus.com/Definitions.aspx?pID=13&lID=57

http://www.simulations-plus.com/Definitions.aspx?pID=13&lID=90

it is an art to make the prediction of permeability from well logs using neural networks because of the lot of considerations.