Diagnosis analytics for energy equipments

linear_algebra

#1

I am stdying an anlytic uses case concerning exceeding energy equipements detection. As ibeginner in data anlytics I am not sure how to begin to resolve this problem.
The use cas is resumed as below :

  • There is a list of 5 equipements. For equipement we measured the activation time for each timestamp.
    For example :
                8h-10h   10h-12h   12h-14h   14h-16h   16h-18h 
equipement 1     100      100       10        0         20 
equipement 2     100      40        10        10        0 
equipement 3     100      32        0         20        0 
equipement 4     0        0         0         100      90 
equipement 5     100      0         0         0         0

And also, I recorded for each timestamp if there in an exceeding energy or not :

         8h-10h       10h-12h     12h-14h          14h-16h      16h-18h 
        exceeding     exceeding   no exceeding     exceeding    no exceeding

My objective in this study is to detect which equipments are responsible for exceeding energy.
Can you propose me some idea to analyze and resolve this problem?

Thank you


#2

For a general answer, you may want to refer to your previous thread: Overcoming energy equipements

Regarding the solution for this particular case, this is more of a linear algebra problem than an analytics one. If you recover the actual numeric values for the output and take the transpose of the equipment-time matrix, this problem can be easily solved with a linear equation solver.


#3

@caiotaniguchi thank you very much for your help. Can you explain me more please. have some tutorial in internet or something that can help me to resolve it? best regards


#4

Tough to say, the answer should be obvious if you know linear algebra. On the other hand, if it’s not obvious, it’s difficult to recommend some material that is not an entire linear algebra course/textbook. Wikipedia might help.

After setting up the appropriate matrix, it’s just a matter of feeding the coefficients to a linear solver. Alternatively, you can apply linear regression instead.


#5

@caiotaniguchi ok thank you, so I need to consider two differents matrix : 1 -

8h-10h 10h-12h 12h-14h 14h-16h 16h-18h
equipement 1 100 100 10 0 20
equipement 2 100 40 10 10 0
equipement 3 100 32 0 20 0
equipement 4 0 0 0 100 90
equipement 5 100 0 0 0 0

and this one 2-

8h-10h 10h-12h 12h-14h 14h-16h 16h-18h
exceeding exceeding no exceeding exceeding no exceeding

I saw the course video of NG in cursera concerning linear algebra, but I dont know to use this in my use case …


#6

Let’s call the equipment-time matrix A^T (A transpose) and the vector b^T. If you can recover the numerical values for b^T instead of using flags, you have a Ax = b kind of problem, which can be uniquely solved with a linear solver, like the one linked in the previous post.

If you cannot recover the values, then the only solution left is to train a classifier like logistic regression, but with only 5 data points I’m not sure it would be of much use.


#7

Ok thank you, is it possible to recover the value by “how much excceding” ?


#8

It should be. I’m assuming these flags are generated after the power consumption exceeds a certain threshold. You just need to find out the true power consumption. As said earlier, flags won’t help you with only this much data.