I'm glad I found these resources. I am an EE with domain knowledge in solar power, and instrumentation background. I would like to get some help determining how I can test the viability of SMILE to detect and diagnose problem in a solar power installation.
THANKS,
-Steve aka PVSleuth
Guidance needed --Diagnosing faults in solar power plants?
-
marek [BayesFusion]
- Site Admin
- Posts: 449
- Joined: Tue Dec 11, 2007 4:24 pm
Re: Guidance needed --Diagnosing faults in solar power plant
Steve,
Thanks for your kind words. Your question is quite general to the point of me not knowing where to start. With a Bayesian network, you can model the joint p.d. over a potentially large number of variables. Once you have a model like this, you can compute the relevance of your observations (e.g., symptoms and test results) to variables of interest (faults). Building the model is a critical part. If you don't have a data set from which you can learn a model, you have to rely on expert knowledge or engineering principles enhanced with expert knowledge. Does all this help at all? If you tell me more about a problem, I can perhaps quickly build a simple model that will illustrate this idea. I propose to move over to email -- my email is marek@sis.pitt.edu.
Cheers,
Marek
Thanks for your kind words. Your question is quite general to the point of me not knowing where to start. With a Bayesian network, you can model the joint p.d. over a potentially large number of variables. Once you have a model like this, you can compute the relevance of your observations (e.g., symptoms and test results) to variables of interest (faults). Building the model is a critical part. If you don't have a data set from which you can learn a model, you have to rely on expert knowledge or engineering principles enhanced with expert knowledge. Does all this help at all? If you tell me more about a problem, I can perhaps quickly build a simple model that will illustrate this idea. I propose to move over to email -- my email is marek@sis.pitt.edu.
Cheers,
Marek
Clarification about Solar power site observables
Dear Marek, all
Sorry about being too vague with my inquiry.
My postulate is that: By using standard measurement in common practice for monitoring a solar power site--Array voltage, currents(strings of modules), irradiance, cell temperature, and sometimes windspeed; updated a few times a minute, and an archive of past 15-minute records we hope to be able to detect faults.
We use a highly abstracted model DC power--
P = Irr * Cos(phi) * D * Area * ( 1- K*dT)
= V * Sum(I's)
see my website www.wattminder.com, to determine benchmark output under current conditions. phi is the incidence angle of Sun's ray with the normal of the solar panel plane. D is the characteristic constant representing the whole collection of modules of a certain brand and model by manufacturer. Area is the total active PV cell area , K is the temperature coefficient per degree C change, dT is the temperature above 25C. Since sensors and data acquisition are expensive, we hope to leverage the power of mathematics and knowledge science to help detect and diagnose for faulty conditions in a solar site. We've been thinking
about Fault dictionary, Conditional Probability Matrix, that relate each observable symptom to every possible fault source. Just hope to understand how we can fit our problem domain into the SMILE and GeNIe frame. Hope this helps you understand our challenge.
Thanks, and May the Sun always warm your face and your panels!
-Steve
Sorry about being too vague with my inquiry.
My postulate is that: By using standard measurement in common practice for monitoring a solar power site--Array voltage, currents(strings of modules), irradiance, cell temperature, and sometimes windspeed; updated a few times a minute, and an archive of past 15-minute records we hope to be able to detect faults.
We use a highly abstracted model DC power--
P = Irr * Cos(phi) * D * Area * ( 1- K*dT)
= V * Sum(I's)
see my website www.wattminder.com, to determine benchmark output under current conditions. phi is the incidence angle of Sun's ray with the normal of the solar panel plane. D is the characteristic constant representing the whole collection of modules of a certain brand and model by manufacturer. Area is the total active PV cell area , K is the temperature coefficient per degree C change, dT is the temperature above 25C. Since sensors and data acquisition are expensive, we hope to leverage the power of mathematics and knowledge science to help detect and diagnose for faulty conditions in a solar site. We've been thinking
about Fault dictionary, Conditional Probability Matrix, that relate each observable symptom to every possible fault source. Just hope to understand how we can fit our problem domain into the SMILE and GeNIe frame. Hope this helps you understand our challenge.
Thanks, and May the Sun always warm your face and your panels!
-Steve
-
marek [BayesFusion]
- Site Admin
- Posts: 449
- Joined: Tue Dec 11, 2007 4:24 pm
Re: Clarification about Solar power site observables
Sorry for missing this. I hope my response is still relevantscyang wrote:Thanks, and May the Sun always warm your face and your panels!
. Interesting wishes -- I actually do have solar panels, so your wishes are well received!It looks like your problem requires continuous variables. I suggest that you build your models using continuous nodes -- you can put equations and distributions in them. The network will compute continuous probability distribution over the domains of the variables in your model. If you run against algorithmic issues, please consider dynamic discretization (one of the tabs in the continuous nodes) -- this should work. If you run against any problems, we can move over to Email: My Email is marek@sis.pitt.edu.
Cheers,
Marek