finish all models learned from the course

Today, I tried all models from the course.

Summary:

5 different models are selected in comparison and the ANN model is the best one.

 Pending tasks:
1. Build the Gaussian Mixture model and compare it with other models.
2. Select the best model and test it in another well.

try OLS and PLS model for sonic prediction

Today, I tried OLS and PLS models for sonic prediction.

Summary:

OLS (ordinary least squares) and PLS (partial least squares) models are to linear regression models with different methods for calculating regression errors for weights and biases.

Tomorrow, I will try other models learned from the course and made a final conclusion for comparison.

finish the first paper

Today, I finished the first paper.

Summary:

I sent you by an email.
Now, we have two papers pending for submission.

Tomorrow, I will continue to compare models for predicting sonic logs. I will try to finish them before holiday.

finish improving the paper

Today, I finished improving the first paper.

Summary:

There are remaining two questions:
1. the editor recommends us to make geology presentation with some graphs. But we do not have. Where can we obtain some schematic graphs about the well location, the formation, and the intervals? If not, I may be just able to plot the schematic graph of intervals.
2. the editor said that our title is not very suitable. Maybe we should change it to another one.

Tomorrow, we can discuss it when we have time and finish the paper.

finish statistical description and performance vs. interval comparison

Today, I finished statistical description and performance vs. interval comparison.

Summary:

I obtained all data for the above two tasks.

Next week, I will start to analyse these data and try to finish the papers ASAP.

think about how to make statistical description and analyze performance related to intervals

Today, I tried to think about how to make statistical description and analyze performance related to intervals.

Summary:

Statistical description:
Compare mean, sd/mean, skewness of every input at different intervals. It needs 3 big tables.
I do not know if it is necessary to plot 3 big tables to do statistical description.

Peformance vs. intervals:
Obtain mean accuracy of every input at different intervals, which needs just one table.

Tomorrow, I will try to finish this part.

finish some parts of improving

Today, I finished some parts of improving.

Summary:

I finished writing the geology part.
I finished replotting the figure for flags and rewriting the paragraph for flags.
I deleted the comparison part between R and Matlab everywhere.

Tomorrow, I will continue to improve the paper following what we discussed today.

improve the paper

Today, I improved the paper by adding the geology section in it.

Summary:

I finished the geology section.
I will change the description of flags, delete some parts of ANN, and write a brief letter for you to the editor as the explanation of our improved work.

Tomorrow, I will try to finish all these tasks.

start to improve the first paper

Today, I started to improve the first paper.

Summary:

I am now adding the section about geology in the paper.

Tomorrow, I will continue to improve the paper and try to finish it ASAP.

finish tasks in outline this week

Today, I finished tasks in outline this week.

Summary:

I analyze distributions and skewnesses of all variables.

Next week, I will see if I should do log change and normalization and skewness change of all inputs and outputs when applying data to models.
Although I took this course and R is used in this course, I still found that it is easier to code in matlab, so i decide to develop all codes in matlab as well.

check the outline

Today, I continued to check the outline.

Summary:

outlier: i should try if outliers will be deleted. if not, we can mention in our paper that we does not delete any data.

PCA analysis should be used for high dimension reduction (thousands or millions), which is not suitable for our research.

transformation of data may be useful for prediction: standardization, change skewness, normalization.

Tomorrow, I will realize codes for this week's outline.

check some tasks in the outline

Today, I checked some tasks in the outline.

Summary:

1. cumulative gains chart and ROC curve are for binary prediction, which is not suitable in our case.
2. boxplots and histograms of data visualization methods may be helpful.
3. pearson correlation coefficient can be used for correlation analysis.
4. concordant and discordant pairs are for value ranking, which is not suitable in our case.

The following is the prediction results with ANN models done before, which is a good prediction with NRMSE less than 0.09.

Tomorrow, I will continue to follow the outline to finish some tasks.

Finish the outline

Today, I finished the outline for sonic log prediction.

Summary:

I will send it to you by an email.

Tomorrow, I will start to finish tasks following the outline.

look for new methods to improve my work

Today, I looked for new methods to improve my work.

Summary:

Today, I looked for new methods and start to write the outline.

Tomorrow, I will finish the outline and start to try these methods.

finish the draft

Today,  I finished the draft for TGRS.

Summary:

I will send you by an email.

Next week, I will try to apply some techniques of the course 'Intelligent Data Analytics' to the prediction of sonic logs.

Finish changing the format

Today, I finished changing the format to TGRS one.

Summary:

I copied all contents of our paper to the template.

Tomorrow, I will edit references, look through the paper, check from top to bottom and send it to you.

continue to write the paper

Today, I continued to write the paper.

Summary:

I continued to write the paper for TGRS.

This weekend, I will try to finish the paper.

continue to write the paper

Today, I continued to write the paper.

Summary:

I am now plotting figures for those results.

Tomorrow,  I will continue to write the paper and try to finish it before next Monday.\

continue to write the paper

Today, I continued to write the paper.

Summary:

I will first finish the draft and then try to satisfy the requirement of TGRS.

Tomorrow, I will continue to write the paper and try to finish it this weekend.

finish the first paper and start on TGRS

Today, I finished the first paper and started on TGRS.

Summary:

I am now writing 'case study' part.

Tomorrow, I will continue to write the paper. I will try to finish this paper this week.

finish the draft

Today, I finished the draft of the first paper.

Summary:

I sent you by an email.

Tomorrow, I will start on the third paper.

finish all calculations

Today, I finished all calculations and comparisons.

Summary:

Good prediction:
porosity: r2=0.7685, nrmse=0.0909;
t2,gm: r2=0.8664, nrmse=0.0840;
t2,gm*porosity^2 (SDR model term): r2=0.7586, nrmse=0.0854.

Bad prediction: BVI, FFI/BVI*porosity (TC model term).

Main reason: when predicting 64 bins, the model consider them as a whole. So the parameter related to all 64 bins are accurate such as porosity and t2,gm, which use all 64 bins for calculation at the same time.
However, BVI and FFI divided T2 distribution into 2 parts, and calculating BVI focuses on just the first several bins. So BVI is not accurate, and FFI/BVI will even increase uncertainty.
As a result, I recommend to mention porosity, t2,gm and SDR model in our validation part instead of mentioning all of them. More research is needed to be done to predict BVI accurately through predicting T2 distribution.
Maybe predicting BVI directly is a good choice.

At the weekend, I will write the paper and finish it. I will send you the draft before Monday.

finish porosity and Swirr comparison

Today, I finished porosity and Swirr comparison.

Summary:

I finished porosity and Swirr comparison and the result is good from the figure.

Tomorrow, I will finish the k comparison and try to finish the paper.

obtain predicted T2

Today, I adjusted models and obtain new predicted T2.

Summary:

The prediction improves a little but almost the same. I saved new T2 distribution and recorded deleted depths. Also, I have changed all data related to the accuracy if changed. So there should be no problem.

Since testing accuracy is more important than training accuracy in data analytics, I use all of testing accuracy when comparing two models. When adjusting the model, narrowing the gap between training and testing accuracy difference is considered.

Tomorrow, I will use new T2 distribution to obtain porosity, k, irreducilbe saturation and bound water in IP. Then I will compare those results  with original ones after depth matching.

calculate irreducible saturation, bound water saturation, k, porosity

Today, I looked for methods to calculate the four parameters.

Summary:

I found that there are some functions in IP software such as NMR interpretation.
I am uploading data to IP to use this function so that I can calculate these parameters.

Tomorrow, I will try to finish this part and make comparison of original and predicted results.

change the first paper and leave some questions.

Today, I changed the first paper and leave some questions.

Summary:

I will send you an email about my questions.

Tomorrow, I will continue to change the paper.

start to write the paper

Today, I started to write the paper.

Summary:

I am now writing the case study part.
I plan to finish the draft next week.

Next week, I will continue to write the paper.

Finish all tasks

Today, I finished all tasks of the paper.

Summary:

Tomorrow, I will start to write the paper.

adjust running the results following the results

Today, I adjusted running the results following the outline.

Summary:

Tomorrow, I will see if I can find more kind of noise and start to write the paper.

add noise to all input logs together and analyze 2 kinds of noise

Today, I added noise to all input logs together and analyzed 2 kinds of noise.

Summary:

Tomorrow, I will see if I can find and analyze other noise and start to write the paper.

add noise to one log at a time

Today, I add noise to one log at a time.

Summary:

I use exactly the same codes for every input but I cannot get prediction results if I add noise to some inputs.

I found that there are some NaNs of the predicted values so that I can not calculate the accuracy. For now, I do not know why it happens.

Tomorrow, I will try to find the reason for the above problem and continue to analyze the noise of logs.

find the least number of inputs

Today, I found the least number of inputs.

Summary:

From the results, we can see that the least number is 12 following the above standard. I deleted RHOZ, PEFZ and GR, which are three least important inputs among all.

Next week, I will start to analyze the noise of inputs and outputs and start to write the paper.

Finish analyzing the dependence of accuracy on each of the inputs

Today, I finished analyzing the dependence of accuracy on each of the inputs.

Summary:

Tomorrow, I will analyze the least number of inputs with which I can still obtain predictions close to the best one now.

search for papers and start to write the paper

Today, I started to search for papers so that I can rewrite the introduction. Also, I list some tasks added to the original work.

Summary:

I am now rewriting the introduction and search for useful papers when writing it.

Also, I listed two more tasks:
1. analyze sensitivity to noise in input data
2. analyze dependence of the accuracy on each of the inputs (rank all inputs using parallel computing)

Tomorrow, I will try to finish the two tasks and continue to write the paper.

Select all useful paragraphs from the second paper

Today, I selected all useful paragraphs from the second paper.

Summary:

I am now writing a draft for the GRSL letter.

Next week, I will continue to write the draft.

start to change the paper

Today, I started to change the second paper to a letter.

Summary:

I am now thinking about how to extract the third model from the whole paper and combine it with inputs analysis.

Tomorrow, I will continue to write the letter.

improve prediction accuracy

Today, I matched the lithology between Atland well and Abernathy well and get better results.

Summary:

Yesterday, I did not match the lithology so that the prediction accuracy is very bad.
Today, I found the problem and matched it to get better results. R2 of Vp and Vs are 0.6419 and 0.5176. NRMSE of Vp and Vs are 0.0928 and 0.0948.
For now, NRMSE is ok but R2 is not high enough (higher than 0.7 is better).
I will see if I can improve it after I finish the letter.

Tomorrow, I will start to think about changing my second paper to a letter.

test models in different wells

Today, I tested models in different wells.

Summary:

I trained the model in Abernathy and test in Atland.
The following are testing results, which is not very good. R2 are -0.6432 and -0.0072. NRMSE are 0.2007 and 0.0593.

Tomorrow, I will check if I can improve the result. If not, I will write the letter first.

finish changing the paper for submission and start to test models in different wells

Today, I finished changing the paper for submission and started to test models in different wells.

Summary:

I sent you by an email.

Tomorrow, I will continue to test models in different wells.

revise the paper and look for reviewers

Today, I revised the paper and looked for reviewers.

Summary:

 Please see the email.

Next week, I will continue the third research.

some journal papers potential for submission

Today, I read some papers related to my research from ATCE. Also , I found some Journals potential for submission.

Summary:

Mathematical Geosciences
https://link.springer.com/journal/11004

Geophysics
http://seg.org/Publications/Journals/Geophysics

IEEE Transactions on Microwave Theory and Techniques
http://ieeexplore.ieee.org/xpl/RecentIssue.jsp?punumber=22

IEEE Antennas and Propagation Magazine
http://ieeexplore.ieee.org/xpl/RecentIssue.jsp?punumber=74

IEEE Transactions on Neural Networks
http://ieeexplore.ieee.org/xpl/RecentIssue.jsp?punumber=72

Tomorrow, we can discuss about which Journal to submit our second paper.

If I can publish one or two papers before application deadline (usually February and March, 2018), it will be very helpful for me to apply for the scholarship. So could you please help me select one Journal for our second paper which will accept our paper with a larger chance and in a higher speed? Thank you so much for your selection and understanding.

recode for ranking

Today, I recoded for ranking all inputs using parallel computing.

Summary:

Next week, I plan to go to ATCE. I will check the model in other 2 wells when I come back.

try to compare resistivity logs

Today, I compared resistivity logs for several times.

Summary:

Tomorrow, I will see if I can find more related to our research.

Finish improvement and ranking

Today, I finished improvement and ranking.

Summary:

Tomorrow, I will read papers and look for some methods to improve prediction performance.

running the improvement flowchart.

Today, I ran the improvement flowchart.

Summary:

Some results are still running. I will upload all of them together tomorrow.

Tomorrow, I will finish the improvement flowchart and rank all inputs.

check data preprocessing methods

Today, I checked all data preprocessing methods.

Summary:

Normalization, PCA, Remove constant values are checked seperately for the ANN model.
Only Normalization helps to train and test the ANN model.
PCA and Remove constant values are used but no improvement is seen.

Tomorrow, I will do other tasks in the list.

realize parallel computing in MATLAB

Today, I realized parallel computing in MATLAB.

Summary:

I checked that Your laptop is 4 core CPU. So theoretically, it will save 3/4 time overall.

I ran 40 times to get the results for Vp and Vs prediction:

Average values: 0.8555 and 0.8295 in terms of R2; 0.0652 and 0.0684 in terms of NRMSE

Best Values: 0.8722 and 0.8492 in terms of R2 and; 0.0599 and 0.0603 in terms of NRMSE

Running time now is 348 seconds (about 900 seconds initially).

Next week, I will finish other tasks discussed today.

try some changes of the parameters

Today, I tried some changes of the parameters of ANN models.

Summary:

1. randomly select initial weights and biases (combine parallel computing to decrease training time)
and record the best result after training ANN models for several times (10 or 20 or 30 ......) I found that the best result among them is a little better than the best result in the first day.
I think it is one effective method to avoid local minimum and obtain global optimization.

2. compare prediction results with and without some data preprocessing steps (such as pca, reciprocal transformation, remove constant values). They can be done after the first step since they may lead to little improvement.

3. another to methods may be useful: Nguyen-Windrow initialization; evaluate effective number of parameters.

Tomorrow, I will validate all of the above methods.

compare three training functions and predict Vp and Vs seperately

Today, I compared three training functions and predicted Vp and Vs seperately.

Summary:

1.
I validate that BR and SCG is equally effective to predict geomechanical data, which are better than LM. So, if we write the paper, we can include them as one of our sections. 

2.
You mentioned that I can try to predict Vp first and Vs next. I did  and I found that prediction accuracy did not improve.
As I mentioned before, accuracy of the ANN model without deleting outliers is 0.8611 and 0.8252 in terms of R2 and 0.0622 and 0.0660 in terms of NRMSE. 
However, when I predict Vp in the first model and predict Vs in the second model, accuracy of ANN models without deleting outliers is 0.8500, 0.8178 in terms of R2 and 0.0710 and 0.0733 in terms of NRMSE.

Tomorrow, I will think about other ways to improve the accuracy.

look for deep learning applications and compare training functions

Today, I looked for deep learning applications and compared training functions.

Summary:

1. I have not found deep learning applications related to our research.
2. I think that training functions may affect the prediction accuracy. I am now comparing LM, SCG and BR. I will select the best one after comparison.

Tomorrow, I will continue above work.

look for applications of deep learning

Today, I looked for applications of deep learning.

Summary:

Deep learning is a type of machine learning, which is well-suited to identification applications such as face recognition, text translation, voice recognition and advanced driver assistance systems.
Usually, there are hundreds of hidden layers of NN models, which needs millions of images and videos to train them. That is why deep learning algorithms can outperform human at classifying images, win against the world's best GO player, or enable a voice-controlled assistant.
However, I have not found applications for function approximation problems, which is like our research problem.

Tomorrow, I will continue to look for deep learning application similar to our research.

finish paper 2

Today, I finished paper 2.

Summary:

I sent you by an email.

Next week, I will improve paper 1 for Fuel.

check all results possible in the basic model.

Today, I checked all results possible in the basic model.

Summary:

I compare 4 conditions.

1. do not delete outliers, ANN model with just one layer
R2: 0.8564 0.8187 NRMSE: 0.0632 0.0672
2. do not delete outliers, ANN model with two layers
R2: 0.8611 0.8252 NRMSE: 0.0622 0.0660
3. delete outliers, ANN model with two layers
R2: 0.8494 0.8265 NRMSE: 0.0695 0.0689
4. delete outliers, ANN model with two layers, predict Vp and Vp/Vs together
R2: 0.8299 0.7964 NRMSE: 0.0712 0.0807

The second performs the best.

Tomorrow, I will continue to look for methods to improve the model such as deep ANN.

finish second paper draft

Today, I finished the second paper draft.

Summary:

I send you by an email.

Tomorrow, we can discuss about which Journal to submit our paper.

Two tables are built.and 4 commons are found

Today, I built two tables for Well 1 and Well 2.

Summary:

two tables are built and 4 commons are found.

1.     Lower mean value of RLA0-5 results in better prediction performance results.

2.     Higher mean value of dielectric dispersion results in better prediction performance results.

3.     Smaller skewness of dielectric dispersion results in better prediction performance results. 

4.     Higher porosity of the formation results in better prediction performance results. 

Tomorrow, I have a project due, I will analyze them on Wednesday. We can discuss if you have any ideas.

finish all except 3.5

Today, I finish all except 3.5.

Summary:

I finish all except 3.5, which is 'petrophysical and statistical controls on the prediction performances'.

Since there are 5 dielectric dispersion, I consider how to use and divide them into several parts of different prediction accuracy.

In the end, I decide to use relative error to measure accuracy of every depth, which is |P-O|/O, P is predicted value, O is original value. For every depth, there are two values, one is the average of conductivity dispersion and the other is the average of permittivity dispersion.

If both of them are less than 0.1, they belong to good prediction performance. If both of them are higher than 0.2 or either of them are higher than 0.3, they belong to poor prediction performance. The rest belong to moderate prediction performance.

Next week, I will try to analyze those results and find some rules.

complete about 80% of the improvement

Today, I complete about 80% of the second paper.

Summary:

I continue to analyze the petrophysical and statistical part.

Since tomorrow is career fair, I will finish the second paper on Friday.

complete about 60% of the improvement

Today, I continue to improve the paper and complete about 60%.

Summary:

I start to analyze the petrophysical and statistical part.

Tomorrow, I will try to finish the paper.

complete about 40% of the improvement

Today, I continue to improve the paper and complete about 40%.

Summary:

I mainly do replotting, recalculation, looking for citations, and adding some parts.

Tomorrow, I will continue to change the paper.

check all wells

Today, I checked all wells to see the difference of accuracy of predicting permittivity dispersion.

Summary:

I checked them one by one and found that they are all worse than the total database of the well.
So I think lithologies at different depths is not the main reason for poor and good prediction performance of permittivity dispersion.

The first figure is NMRSE result of well 1 (6 different lithologies).

The second figure is R2 result of well 1 (6 different lithologies).

Next week, we can discuss about our second paper when you have time.

finish the final draft of the second paper

Today, I finished the draft of the second paper.

Summary:

I have checked different lithologies one by one in well 1  and find that all results are worse than the total one, so i think lithologies at different depths may not be same main reason.

I conclude 2 reasons in our paper:

1. fewer data samples
2. high water salinity

Details can be seen in the draft that I sent to you.

Tomorrow, maybe we can discuss which Journal we can submit our paper to and change the format for submitting. Also, we can continue to improve our paper if you have more suggestions.

finish applying models

Today, I finished applying models to different wells.

Summary:

I finished applying models to different wells.

I am now analyzing some problems:
1. why prediction performance in well 3 is bad?
2. why prediction of cond is always better than perm?
3. why NRMSE is better than R2 for estimation of accuracy?

Tomorrow, I will try to finish all the analyzing parts.

do the second task left last week and finish half of them

Today, I did the second task left last week and finished half of them.

 

Summary:

I trained and tested the third method in well 1 and apply it in well 2. I will finish the opposite tomorrow.

Tomorrow, I will continue to change the second paper.

leave two tasks

Today, I improved all other parts except two tasks.

Summary:

Two tasks:

1. Evaluation of NRMSE and R2 and their explanation

2. Train and test the third method in well 1 and apply it in well 2 and vice versa

Next week, I will try to find good explanation of NRMSE and R2 and realize the application of the third method in 2 wells.

continue to improve the second paper

Today, I continued to improve the second paper.

Summary:

I finish all improvement before 'case study'. There are now two parts left to be improved, which are 'case study' and 'conclusion'.

Tomorrow, I will try to finish all the rest part.

Improve the second paper

Today, I improved part of the second paper.

Summary:

I will try to finish it on Thursday or Friday.

Tomorrow, I will continue to improve the second paper.

some ideas after reading papers combined with your suggestions

Today, I thought of some ideas after reading papers combined with your suggestions.

Summary:

1. CMIS: in other papers, people combine different models into CMIS like fuzzy-logic, neuro-fuzzy and ANN. Since ANN performs the best among selected models, I think one improvement may be to use different ANN models in CMIS instead of various models. For now, I have not found how to realize it in MATLAB. I will continue to look for it.

2. one way to use Vp/Vs is to predict Vp and Vp/Vs together instead of predicting Vp and Vs together. From yesterday's work, it is more accurate to predict Vp than Vs, so predicting Vp and Vp/Vs together may be a good choice.

Tomorrow, I will start to improve the second paper.

predict sonic logs with basic ANN models

Today, I predicted sonic logs with basic ANN models.

Summary:

There are 13 inputs and 2 outputs (compressional and shear).

one hidden layer with 8 neurons
R2: 0.8570, 0.8184
NRMSE: 0.0631, 0.0673

two hidden layers with 9 and 5 neurons
R2: 0.8619, 0.8247
NRMSE: 0.0620, 0.0661

Asoodeh et al. (2011) improved accuracy using CMIS compared with ANN. But it is just 0.16% for compressional velocity and 0.63% for shear velocity with R2.

In our case, the accuracy is also improved by adding one more hidden layers. It is 0.57% for compressional velocity and 0.77% for shear velocity with R2.

The result shows that both of two methods improve little in accuracy and there is no need to use CMIS. Adding one more hidden layer is more efficient.

Tomorrow, I will search papers to see if I can find new methods.

Possible improvement from former research

Today, I thought about possible improvement from former research about predicting sonic logs.

Summary:

1. from papers I read yesterday, most predictive models use Vp as inputs with other conventional logs to predict Vs. Most empirical equations predict Vp and Vs together with limited accuracy.

2. There is one paper where Vp and Vs are predicted together with CMIS. There are three different models in CMIS including ANN, FL (fuzzy logic) and NF (Neuro-Fuzzy) models. The accuracy of CMIS is a little higher than ANN and much higher than FL and NF. So maybe there is no need to  use CMIS.

Possible improvements:

1. Just use ANN to predict Vp and Vs together instead of CMIS

2. improve the performance of ANN and obtain higher accuracy than CMIS with less time (contain more inputs, more hidden layers, predict one of Vs and Vp first and predict the other next to get higher accuracy, select the best training function)

Next week, I will see if it is possible to use ANN to obtain higher accuracy than CMIS.

literature review of sonic logs

Today, I did literature review of sonic logs.

Summary:

There are two main kinds of methods of predicting Vp and Vs.

1. data from lab measurements
1.1. multiple regression
1.1.1. Inputs: NPHI, RHOB, GR, LLD(deep laterolog, resistivity), Vp
          Outputs: Vs
1.1.2. Inputs: NPHI, RHOB, Vp
          Outputs: Vs
1.2. ANN
Inputs: NPHI, RHOB, Vp
Outputs: Vs

2. data from well logs
Xu-White model

In conclusion,  following logs can be selected as inputs and outputs for our research.
Inputs: GR, AT10, AT90, DPHZ, NPOR, PEFZ, RHOZ, VCL
Outputs: DTCO, DTSM

In the evening, more literature review are done for predicting sonic logs.

In Oregon Basin, Wyoming, Iverson and Walker (1992), empirical equations
inputs: shale content (from GR), lithology and porosity (nuclear logs)
outputs: shear and compressional transit times

sand/shale mixture, Dvorkin and Gutierrez (2001), empirical equations
inputs: mineral properties, rock porosity and shale volume
outputs: Vp and Vs

South Texas clastics, offshore gulf of mexico clastics, Greenberg and Castagna (1992), empirical equations
inputs: lothology, porosity, water saturation, Vp
outputs: Vs

Asmari formation (carbonate reservoir rock of Iranian oil field), Asoodeh (2011), CMIS (committee machine with intelligent sustems)
inputs: NPHI, RHOB, Rt, Vsh
outputs:Vp, Vs, stoneley wave velocity

sandstone reservoir of Carnarvon Basin, Australia, Rezaee et al. (2006), CMIS (committee machine with intelligent sustems)
inputs: NPHI, Vp, GR, RHOB, Rlld
outputs: Vs

carbonate reservoir in Iran, Maleki et al. (2014), empirical correlations and AI methods (SVM, ANN)
inputs: Vp, RHOZ, GR, PHIT, Rt, HCAL
outputs: Vs

Tomorrow, I will start to build ANN for DTCO and DTSM.

finish some tasks

Today, I finished some tasks.

Summary:

1. finish compile BHP and Hess dataset.

2. check that VPVS=DTSM/DTCO
    I check it in Hess well and it is true. I did not find VPVS in BHP well. I think it also means that           VPVS is obtained by DTSM/DTCO.

3. do some research on sonic logs. There are some applications for them:
    quality control of transit-time logs
    refinement of porosity predictions
    determination of lithology
    improvement of fracture detection
    improvement of cement bond evaluation

Tomorrow, I will learn more about sonic logs and start to predict them by conventional logs.
   

prepare the compilation of Hess well

Today, I prepared the compilation of Hess well.

Summary:

I have not finished it yet.

Tomorrow, I will try to finish both the Hess well and the BHP well.

make a conclusion of the past work and think about the next step

Today, I made a conclusion of the past work and wrote my SOP. In addition, I looked into RV and RH.

Summary:

Since the application deadline of Australian universities for next Fall is in October, I have to accelerate preparing for the application materials.

In addition, I considered about predicting RH and RV. It is possible because resistivity reflects the characteristics of both formation rocks and fluids, which are related to various logs. However, since resistivity logs are not quite expensive, maybe there is little application if the research can be done.

Tomorrow, we can talk about this as well as my next step of my research.

Finish the draft of the second paper

Today, I finished the draft of the second paper.

Summary:

It is in my email.

Next week, I will do research on RV/RH.

Finish part 2 of the second paper

Today, I finished part 2 of the second paper. I am now writing part 3 (case study), which is the only part left.

Summary:

I will send you an email.

Tomorrow, I will try to finish the draft.

continue to write the second paper

Today, I continued to write the second paper including some of the abstract, introduction and methodology. I also changed the flowchart of the third method.

Summary:

It is in my email.

Tomorrow, I will continue to write the second paper.

finish the instruction and start writing other parts

Today, I finished the instruction and started writing other parts of the second paper.

Summary:

It is in my email.

Tomorrow, I will write the following parts.

Write introduction of the second paper

Today, I wrote introduction of the second paper.

Summary:

It is in my email attachment.

Tomorrow, I will finish the introduction and continue to write other parts.

finish changing the paper to Fuel version

Today, I finish change the paper to Fuel version.

Summary:

I will send you an email.

Tomorrow, I will improve it if you have any instructions.

finish the first paper

Today, I finish the first paper and write a little of the second paper.

Summary:

I will send you an email.

Tomorrow, I will select one journal for the submission of the first paper.

improve all figures and some contents of paper 1

Today, I improve all figures and some contents of paper 1.

Summary:

I will send you an email.

At weekends, I will finish improving paper 1.

finish writing conclusions of the second paper

Today, I finish writing conclusions of the second paper.

Summary;

I will send you by an email.

Tomorrow, I will improve the first paper following your instructions.

plot log data and build three models for atland well

Today, I plot log data and build three models for atland well.

Summary:

I will send you by an email.

Tomorrow, I will continue to write the second paper.

build the third model for abernathy well

Today, I compared two methods to build the third model.

Summary:

method 1: train all 8 models with log data and original dielectric data as inputs, test all 8 models with log data and predicted dielectric data as inputs.

method 2: train and test all 8 models one by one with log data as predicted dielectric data as inputs.

After comparison, it shows that method 1 has much worse accuracy than method 2 does. So method 2 should be used to build the third model.

I will send you by an email.

Tomorrow, I will start to write the second paper since I have all results of figures and tables now.

replot the figures and build method 2

Today, I replotted the figures and build method 2

Summary:

I will send you by the email.

Next week, I will build method 3 and start to write the second paper.

plot figures and tables in the outline

Today, I plotted figures and tables in the outline.

Summary:

I will send you by the email.

Tomorrow, I will finish plotting figures and tables for one well in three methods.

design the distribution examples of flags

Today, I designed the distribution examples of flags.

Summary:

I will send you by the email.

Tomorrow, I will continue to build the three models for one well.

finish preparing for the results of the second paper

Today, I finished preparing for the results of the second paper.

Summary:

I will send you by the email.

Tomorrow, I may start to write the second paper.

finish the task of abernathy well

Today, I follow your instructions to compare the three wells and finish the abernathy well.

Summary:

I have finished the abernathy well. I will send it to your by email.

Tomorrow, I will finish compare the rest two wells.

do some data preprocessing to improve the model

Today, I did not figure out how to improve the model. I just delete 66/4359 outliers to obtain a slightly better result.

Summary:

First, I changed the order of predicting these 8 outputs. I changed the order of perm 4 and perm 3 so that both orders of cond and perm are 2134, which is more reasonable. 8 changed models are shown as follows.

ANN1: log data   to   cond 2;
ANN2: log data + cond 2   to   cond 1;
ANN3: log data + cond 2 1   to   cond 3;
ANN4: log data + cond 2 1 3   to   cond 4;
ANN5: log data + cond 2 1 3 4   to   perm 2;
ANN6: log data + cond 2 1 3 4 + perm 2  to   perm 1;
ANN7: log data + cond 2 1 3 4 + perm 2 1   to   perm 3;
ANN8: log data + cond 2 1 3 4 + perm 2 1 3   to   perm 4.

The result improved 8 R2 of testing data in the above order are 0.91, 0.92, 0.88, 0.87, 0.71, 0.70, 0.65, 0.61 in the above order.

5 of them are higher, 2 of them are the same and just 1 is lower, which is perm 4.

For now, I predict 1 output and add predicted values into next model's inputs to train and test the next model. I plan to build all eight models with original data first and use predicted data to calculate the outputs directly next week. I think it may help.

Next week, I plan to do as the above idea.

predict 8 outputs one by one to get better results

Today, I predicted 8 outputs one by one to get a better result than predicting them together.

Summary:

The following two methods are inspired from honoring relationship between 4 cond and 4 perm.

First, I build two models. The first is to use log data to predict  4 cond. The second is to use 4 cond to predict 4 perm.
The prediction accuracy of 4 cond is good, 4 R2 of testing data are 0.80, 0.90, 0.86, 0.86.
But using 4 predicted cond to predict 4 perm is not good. 4 R2 of testing data are 0.54, 0.65, 0.56, 0.47.

Second, I build eight models.
I predict 8 outputs together first for ten times and average the R2 results. The accuracy from top to bottom is cond 2 1 3 4 and perm 2 1 4 3.
ANN1: log data   to   cond 2;
ANN2: log data + cond 2   to   cond 1;
ANN3: log data + cond 2 1   to   cond 3;
ANN4: log data + cond 2 1 3   to   cond 4;
ANN5: log data + cond 2 1 3 4   to   perm 2;
ANN6: log data + cond 2 1 3 4 + perm 2  to   perm 1;
ANN7: log data + cond 2 1 3 4 + perm 2 1   to   perm 4;
ANN8: log data + cond 2 1 3 4 + perm 2 1 4   to   perm 3.

The result improved. 8 R2 of testing data in the above order are 0.89, 0.91, 0.88, 0.87, 0.71, 0.64, 0.64, 0.65.

Tomorow, I will see if I can improve prediction accuracy by other methods.

read papers about cole-cole equation and debye equation

Today, I read papers about cole-cole equation and debye equation.

Smuuary:

cole-cole equation and debye equation are highly related.
The first equation is cole-cole equation. the second is debye equation, when alpha=0, cole-cole equation becomes debye equation.
the third is the real part permittivity equation, the fourth is the imaginary part permittivity equation. they are all functions of w.

I think i can obtain all other constants of the third and fourth equations so that I can know the relationship between real part permittivity and w, imaginary part permittivity and w. They may help me improve the prediction accuracy.

Tomorrow, I will try and validate this method.

Finish changing the paper

Today, I finished changing the first paper.

Summary:

I emailed to you with the draft.

Tomorrow, I will continue to focus on my second research.

Improve the first paper

Today, I improved the first paper following Yifu's instructions.

Summary:

I changed part of the paper. I will finish it tomorrow. Yifu told me to add some contents in the paper and change all colored figures into black-and-white ones, which is time-consuming.

Tomorrow, I will continue to change the paper.

think about the physical relationship between cond and perm

Today, I think about the physical relationship between cond and perm and why it is more accurate to predict magnitude and phase than cond and perm.

Summary:

 Next week, I will see if I can find a better explanation.

verify the method in 'at land state' well

Today, I verify the method in 'at land state' well.

Summary:

I also changed original cond and perm into magnitude and phase in 'at land state' well to get a good prediction result.

The first is the prediction performance of magnitude. R2 of testing data are 0.91, 0.87, 0.80, 0.68.

The second is the prediction performance of phase. R2 of testing data are 0.87, 0.93, 0.88, 0.82.

Tomorrow, I will compare all inputs to know better about their effects on prediction performance.

Use magnitude and phase to improve the result

Today, I read some papers and decide to use magnitude and phase again.

Summary:

There are some papers which predict magnitude and phase of one value like permittivity. So I think it may be helpful. After applying this method into abernathy data, the accuracy of prediction improves a lot.

The first is the prediction result for magnitude. R2 of testing data are 0.93, 0.91, 0.88, 0.77.

The second is the prediction result for phase. R2 of the testing data are 0.85, 0.94, 0.91, 0.88.

 I think the accuracy now is enough for publication.

Tomorrow, I will continue to read papers if you can give me some time.

read some papers

Today, I read some papers, but I cannot find something helpful to use the relationship among frequency, conductivity and permittivity creatively.

Summary:

one method to predict permittivity is:
1. simulate magnitude and phase with FDTD (imputs are real and imaginary part of permittivity)
2. use magnitude and phase as inputs and permittivity as outputs to train the ANN model
3. measure other magnitudes and phases of different materials
4. use data from step 3 and model from step 2 to predict permttivity

another method to predict permittivity is:
1. obtain inputs from elementary measurements
2. build functions of permittivity and locations (linear, quadratic and Gaussian function) for different samples
3. train an ANN model with locations as inputs and parameters of functions as outputs.
4. input other locations to the model and predict permittivity
5. obtain 2-D complex permittivity profiles

FDTD (Finite-difference time-domain method) is a numerical analysis technique used for modeling computational electrodynamics.

select some depths and get better results

Today, I select some depths and get better results.

Summary:

There are 6731 depths I use as training and testing data last week. They are from 6890 ft to 10255 ft. Today, I tried several times and select depths from 7840 ft to 10019 ft. They are more similar to each other according to lithology. Actually, I get better results from them.

The following is the prediction result of 4 con. They are predicted by 15 logging data. R2 of 4 con are 0.91, 0.90, 0.88 and 0.85, which are very high.

The following is the prediction result of 4 per. They are predicted by 15 logging data and predicted 4 con just now. R2 of 4 per are 0.66, 0.69, 0.61 and 0.62. Although they are not as high as 4 con's. But they are better than last week's results.

Tomorrow, I will try other methods to improve the prediction performance.

continue to change the paper

Today, I continue to change the paper.

Summary:

I will send you by email once I finish today.

Next week, I will focus on improving the prediction performance of abernathy well.

change the first paper

Today, I change the first paper.

Summary:

I spend all day to change the first paper but I has not finished it yet. I will try to finish it tomorrow.

Tomorrow, I will finish changing the paper and send it to you.

apply the model in abernathy well and get better results

Today, I apply the model in abernathy well of BHP data and get better results.

Summary:

I discussed with Yifu. We conclude that because of high water salinity in bakken formation, the permittivity values are extremely large. But there are no logging data which can reflect this characteristic. However, the water salinity in permian basin is not high, so the prediction in abernathy well is much more accurate.

Here are the prediction performance results of abernathy well.

The first is to use original 4 con and 15 inputs to predict 4 per directly. R2 of 4 per of testing data are 0.91, 0.92, 0.80, 0.66. They  are much better than that in Hess data. It shows that it is feasible to predict 4 con first and use it with logging data together to predict 4 per. But the problem is that the accuracy varies a lot.

The second is to predict 4 con first and then use predicted 4 con and logging data together to predict 4 per. At first, I predict 4 con. R2 of 4 con of testing data are 0.88, 0.87, 0.85, 0.82.

Then 4 per are predicted by 15 inputs and predicted 4 con in another ANN model. R2 of 4 per of testing data are 0.61, 0.65, 0.55, 0.51. They are more accuracy and stable than those in Hess data.

Tomorrow, I will continue to improve the prediction performance. In addition, Hao has given me part of the suggestions of improving the paper just now. I will improve it tomorrow.

Predict magnitude and tan

Today, I predict magnitude and tan. The result is not very good.

Summary:

First, I turn 4 con and 4 per into 4 magnitude and 4 tan values. The I predict them together. The results of prediction performance are shown as follows.

The first is four magnitude. The R2 of four magnitude prediction of testing data are 0.78, 0.75, 0.71, 0.77, which are almost as good as 4 con prediction.

The second is four tan value. The R2 of four tan value prediction of testing data are 0.23, 0.24, 0.22, 0.58, which are almost as bad as 4 per prediction.

In conclusion, for now, they best method is to predict 4 con first and then use 4 con and 11 log inputs to predict 4 per.

Tomorrow, I will continue to try other methods.

add predicted conductivity to inputs ang read more papers

Today, I add 4 predicted conductivity into 11 inputs to get better results. Also, I read some papers about how to predict dielectric data with ANN.

Summary:

If predicted conductivity are added into inputs, the prediction performance of permittivity will be improved. The following is the result of predicting permittivity. All R2 are improved. The 4 R2 of permittivity prediction are 0.12, 0.24, 0.44, 0.69. The last permittivity with the highest frequency has reached to the ideal accuracy.




In addition, I read some papers about how to predict dielectric data with ANN. For now, I have not found methods which are useful for my research.

Tomorrow, I will continue to read papers and try different methods.

check lithologies and try shallow data

Today, I check lithologies.

Summary:

There are ten different lithologies contained in total, from 2 to 11. After trial and error, the best is to use lithologies from 3 to 11. I also deleted the last 15 depths for better data.

The results are as follows. The first is 4 conductivity prediction. Four corresponding R2 of testing data are 0.76, 0.77, 0.69, 0.77. They are a little better than yesterday's results.

The second is 4 permittivity prediction. Four corresponding R2 of testing data are 0.05, 0.28, 0.35, 0.61. They are a little better than yesterday's results.

In conclusion, the lithologies have some kind of effect on prediction performance but not much. For now, lithologies from 3 to 11 together can predict dielectric data the best.

In addition, I also checked shallow data (what we use before are deep data). But the result shows that they are worse than deep data. I checked the all of them in IP and found that shallow data are even more in disorder than deep data. That may be the reason why shallow data have worse prediction performance.

In conclusion, we should still use deep data.

 Next week, I will check the following things which could be useful for improving prediction performance:
1. change the training function and change the model;
2. use predicted con with 11 inputs to predict per together;
3. produce magnitude, tan and predict them;
4. predict with NMR.

Follow the instructions

Today, I follow the instructions as the meeting.

Summary:

First, I predict 4 real con and 4 imaginary con together with 19 inputs. The results of 4 imaginary con are still not good, which are converted from real permittivity.
The reason may be that the conversion is just multiplying some parameters. So there will be little help to improve the prediction.
The first figure is 4 real con prediction performance. The R2 of the testing data are 0.76, 0.83, 0.63, 0.77, which are as good as before.

 The second figure is 4 imaginary con prediction performance. The R2 of the testing data are 0.14, 0.17, 0.29, 0.54, which are as bad as before.

In conclusion, the conversion from real permittivity to imaginary conductivity may be little help for improving the prediction performance.

Second is to predict them with just 11 inputs.
The first figure is 4 real con prediction performance. The R2 of the testing data are 0.75, 0.74, 0.73, 0.74.

  The second figure is 4 imaginary con prediction performance. The R2 of the testing data are 0.05, 0.16, 0.42, 0.52.

In conclusion, the deletion of 8 inputs make the prediction of 4 real con more stable and increase the prediction of 4 imaginary con a little. Generally, 11 inputs are similar to 19 inputs in prediction performance. Actually, more inputs will not decrease the prediction performance. However, we can use just 11 inputs to predict 8 con since 19 inputs do not perform much better.

Third, I check different lithologies step by step. For now, there are 10 different lithologies before division, category from 2 to 11.

1. I check the performance by selecting category from 4 to 10.
The first  is 4 real con prediction performance. The R2 of the testing data are 0.68, 0.58, 0.62, 0.66.
The second  is 4 imaginary con prediction performance. The R2 of the testing data are 0.13, 0.15, 0.32, 0.39.
So the performance is worse.

Tomorrow, I will continue to check lithologies. After that I will continue the instructions with magnitude and tan and NMR. Also, the change of training function may be a choice.

predict Sw

Today, I continue to read some papers and start to predict Sw.

Summary:

Some properties which may affect dielectric logging data:
the content of water, oil and rocks;
different rock minerals;
different lithologies;
mud cake;
water salinity;
temperature, so maybe depth;
in low frequency, conductivity dominates, in high frequency, dielectric constant dominates

Since we use permittivity (both dielectric constant and conductivity) to calculate Sw with some mixing models as I read from papers, why not predict Sw directly?
I found Sw in dielectric data and I predict it.

The training R2 is 0.75 however the testing R2 is 0.07, which is very bad.
I think I can first try to find methods to predict Sw of dielectric data accurately. Then maybe I can also predict per and con more accurately.

Tomorrow, I will try to improve the performance of prediction for Sw.

Finish reading about basic knowledge of permittivity and conductivity

Today, I read three more papers and know about dielectric data more.

Summary:

The task of quantifying petrophysical parameters is divided into two steps.

1.      Invert for the electromagnetic properties of the formation (permittivity and conductivity) from magnitude and phase of electromagnetic wave recorded at the receiver. These electromagnetic properties are obtained for each frequency of operation to obtain their variation with respect to frequency (dispersion).

2.      Valid mixing models are required to relate petrophysical parameters of formation and dispersive electromagnetic properties.

It is shown that the presence of noise in the data can lead to ill-posed inverse problems where multiple answers can be present.

Monte Carlo method is used to study the effect of the noise on the inverted petrophysical properties of the formation.

With the increase of salinity in water, the permittivity of water decreases a little while the conductivity of water increases much. In addition, water salinity will not affect the properties of rocks much.

There are many methods developed for dielectric data. Some of them are for inverting for permittivity and conductivity with different frequencies in order to obtain petrophysical properties. Some of them are for defining quality control criteria for dielectric logging.

Tomorrow, I will start to look for methods to improve the prediction performance of dielectric data. Also, I will try to find papers about how to predict dielectric data accurately.

there is not a single relationship between permittivity and frequency

Today, I read two more papers and find something new.

Summary:

Tomorrow, I will try to finish reading and start coding. However, I really want to spend a little more time on reading so that I can know more about dielectric dispersion.

Basic knowledge of permittivity

Today, I read some papers to better understand permittivity and conductivity.

Summary:

On Tuesday and Wednesday, I met some friends in SPWLA and learn a lot from them. One of them also does research on dielectric data and he sent me some papers for me to better understand dielectric data.

Next week, I will continue to learn about the basic knowledge of dielectric data and try to find solutions for predicting them accurately.