User:Jaired

From REU@MU
Revision as of 17:19, 9 July 2018 by Jaired (Talk | contribs)

Jump to: navigation, search

I am Jaired Collins, a student at Missouri Southern State University in Joplin, MO, currently majoring in computational mathematics. Outside of academia, I enjoy cycling, weights, table tennis, tennis, and cooking.

Week 1 (29 May - 1 Jun)

Tuesday

  • Initial meeting to get introduced into the program.
  • Lunch meeting with mentors.
  • Obtained MU and MSCS account logins and ID card
  • Paperwork

Wednesday

  • Met with Dr. Corliss to get acquainted with the GasDay lab and receive a key.
  • Met with Dr. Povinelli and another student to set up initial parameters of the project.
  • Literature review.

Thursday

  • More literature review
  • Seminar with GasDay
  • Meeting with Dr. Povinelli and student to establish a weekly plan and talk about the project.

Friday

  • Literature review
  • Planned for a presentation at a GasDay Seminar on the 21st of June
  • Data cleaning


Week 2 (4 Jun - 8 Jun)

Monday

  • Meeting with Dr. Corliss
  • Meeting with Dr. Povinelli and Colin
  • Literature search
  • Cleaned up more data

Tuesday

  • Meeting with Dr. Corliss
  • Data cleaning with bash and Matlab
  • Played around with data and SVMs in MatLab
  • Meeting with research consultant
  • Literature review

Wednesday

  • Cross validation with k-Fold
  • GasDay Camp
  • Literature search

Thursday

  • More built-in SVM work in MatLab
  • GasDay camp
  • REU working lunch meeting, bad presentation

Friday

  • Meeting with Dr. Corliss
  • Creation of ShareLaTeX project
  • Converted temporal pressure data into nonuniform time series data
  • Tried resampling time series data into uniform data


Week 3 (11 Jun - 15 Jun)

Monday

  • Meeting with Dr. Corliss
  • Meeting with Dr. Povinelli
  • Resampled nonuniform data
  • Forecasting with ARIMA
  • Left computer running to finish ARIMA models

Tuesday

  • Examined and saved ARIMA models
  • Dissimilarity-based approach to prediction, where correlation coefficients between sensors are used as features
  • SVM and k-fold cross validation at k = 5, 10, and 20

Wednesday

  • Meeting with Dr. Corliss to elucidate LS-SVM for regression
  • Coding LS-SVM
  • RCR lunch meeting

Thursday

  • Meeting with Dr. Corliss, further explained LS-SVM
  • GasDay seminar
  • Coding of LS-SVM for regression
  • Meeting with Dr. Povinelli
  • Called contact to clarify threshold values

Friday

  • Out for wedding


Week 4 (18 Jun - 22 Jun)

Monday

  • Out for wedding

Tuesday

  • Wrote working abstract
  • Wrote summaries for related works section
  • Read Anomaly Detection in Streaming Non-stationary Temporal Data
  • Read Chapter 5 of Operations & Maintenance Best Practices Release 3.0
  • Tested different features in LS-SVM; residuals are very low

Wednesday

  • Meeting with Dr. Corliss, showed LS-SVM results
  • Work on conference paper with graduate student
  • LS-SVM Tuning
  • Work on PPT for presentation

Thursday

  • Practice presentation at GasDay Seminar
  • Working REU lunch
  • Knowledge and Information Discovery with Dr. Povinelli
  • Project meeting with Dr. Povinelli and graduate student

Friday

  • Recursive nonlinear autoregressive model with LS-SVM
  • Clarification on building different models with different time horizons instead of recursive approach
  • Coding nonlinear autoregressive model with LS-SVM with multiple time horizons
  • Work on abstract
  • Work on background


Week 5 (25 Jun - 29 Jun)

Monday

  • Worked on slides for presentation on Thursday
  • Tried to debug direct multiple time horizon least squares support vector machine nonlinear autoregressive model (LS-SVM NAR)
  • Showed abstract to Dr. Povinelli, got feedback
  • Resampled data to 1 minute and wrote several MATLAB scripts to automate the LS-SVM NAR process
  • Came to the conclusion that LS-SVM NAR is not predicting well at all. Plotted values were erroneous and coincidentally showed promising results

Tuesday

  • Meeting with Dr. Corliss
  • Read more into SVMs and SVRs to help the LS-SVM NAR algorithm
  • Implementation of radial basis function kernel for LS-SVM NAR

Wednesday

  • Debugging LS-LSVM NAR with RBF kernel with test cases
  • Tuning LS-LSVM NAR hyperparameters
  • Presentation preparation

Thursday

  • More preparation for REU Midterm
  • GasDay Seminar
  • Midterm presentations
  • KID Seminar
  • Meeting with Dr. Povinelli to discuss project

Friday

  • Put a lot of effort into making LS-SVM work
  • Meeting with Dr. Povinelli to help with LS-SVM and things to do while he is on vacation


Week 6 (2 Jul - 6 Jul)

Monday

  • Confirmed some math with Dr. Corliss
  • Messing around with LS-SVM to get it to work
  • Created normalized time series
  • Extracted correlated information from multiple sensors

Tuesday

  • Implemented long short-term memory recurrent neural network, peaks too often
  • Tried several configurations of LS-SVM NAR
  • Created "artificial" data with extra troughs in it to improve accuracy of troughed data with LS-SVM
  • Literature search on time series/forecast*/machine learning

Wednesday

  • Created "artificial" data with extra peaks in it to improve accuracy of peaked data with LS-SVM
  • Manual ensemble of peaked, trough, and normal LS-SVM models for forecasting
  • Created several figures to show LS-SVM for regression

Thursday

  • Remote desktop into a better computer to use more training data
  • Started working on a programmatic rule-based ensemble of differently trained models
  • REU working lunch with Dr. Corliss


Week 7 (9 Jul - 13 Jul)

Monday

  • Meeting with Dr. Povinelli to discuss project and his and graduate student's visit to the customer company
  • Created different figure of LS-SVM to only include the 30 minute forecast value every minute
  • Analyzed all sensor information from all the sites to find anomalies


Week 8 (16 Jul - 20 Jul)

More experiments with algorithm

Week 9 (23 Jul - 27 Jul)

Work on final draft of research paper
Work on poster

Week 10 (30 Jul - 3 Aug)

Prepare final presentation