C31 Optimizing a Continuous Process (Kay)Title: C31 Optimizing a Continuous Process
Author: Phil Kay
Enclosed: I have a demo for adding to the technical reference library. It is on a continuous process. There is a data table and a brief slide to introduce the situation.
Why Good: I think it is a useful demo for predictive modeling and JMP Pro, especially in a chemicals or biotech situation.
The system behaviour is non-linear. There are many variables and unknown time lags between each variable and the response. Finding behaviours with standard time series analysis will be very challenging. So the approach is to create a range of lag variables for each original variable and then throw them all into a BF or NN. In JMP13 you can create lag these variables very easily.
Attachments:
- Continuous Process 100628.pptx
- Continous process data log.jmp
Case StudiesRe: C31 Optimizing a Continuous Process (Kay)There are loads of time series datasets here:
https://archive.ics.uci.edu/ml/datasets.php?format=&task=&att=&area=&numAtt=&numIns=&type=ts&sort=nameUp&view=table
Perhaps one of them would help.
I think I have used the "Ozone Level Detection" data set before. But I don't remember much about it other than it was interesting enough to use for a general demo.
Re: C31 Optimizing a Continuous Process (Kay)Hi @alisa_h_lowery
I am not aware of any non-manufacturing case study data sets like this. Sorry.
I would think that there would be quite a lot out there as time series data analysis is very common outside of science and engineering (finance, retail, etc.).
I couldn't find any examples that were suitable for process industries and so that is why created this.
Phil
Re: C31 Optimizing a Continuous Process (Kay)There is a weekly weather data example in FDE that shows how the Fourier Basis Functions and fPCA are used to extract features.
Although not a key application area, SEM can be used to model multivariate time series. Laura Castro-Schilo shows how a VAR (Vector Autoregressive Regression) model can be fit with SEM. https://community.jmp.com/t5/Discovery-Summit-Europe-2021/Structural-Equation-Modeling-of-Coupled-Twin-Distillation/ta-p/349223?_ga=2.241780289.218951656.1617286921-1157223820.1580417171
You can try it on the GNP.jmp in the JMP sample data library. I haven’t done that yet, so I don’t know how the SEM results compare to Proc VARMAX in SAS/ETS.
Re: C31 Optimizing a Continuous Process (Kay)@Phil_Kay @jiancao Are there any similar examples for non-manufacturing that showcase using Pro to demo Time Series data?
Re: C31 Optimizing a Continuous Process (Kay)Cool. The advantage of this lag variable + machine learning approach over standard time series analysis is that it can find nonlinear relationships between variables. This demo was created to sell JMP Pro. I'll maybe add something like that in the description.
Re: C31 Optimizing a Continuous Process (Kay)Hi @Phil_Kay
No need of your time but thank for the proposal. I think I catch the spirit of the demo looking at the data. The funny thing is that it is somethning I was looking for (how to evaluate the lag time on a continuous process) but I found it while looking at something else (time series). Thank you Phil
Emmanuel
Re: C31 Optimizing a Continuous Process (Kay)Hi @Emmanuel_Romeu ,
I have deleted the reference to the webex video as I am fairly sure that is lost now - sorry. In any case, the video was only intended for internal viewers - it was not suitable for customers. I am not aware of another recording of this demo.
I am happy to talk through the demo with you if you like though. I think @Florian_Vogt may have used this case study recently.
Cheers,
Phil
Re: C31 Optimizing a Continuous Process (Kay)Hello,
the link to the video does not work.
I wanted to share the recording with a customer that's maybe not possible as located in Inside JMP ?
@Phil_Kay @scwise
Emmanuel
