A step by step exercise of how to do probabilistic planning 3

[andrewward]

Hi All,

I have been dropped into the deep end of reliability analysis and probabilistic planning for power systems (transmission and distribution) and need to get up to speed very quickly.

I have read several IEEE papers on the topic and am now starting to understand monte carlo analysis and also the analytical methods. E.g. I understand the need for outage data (or monte carlo in the absence of outage data), cost of non-supply data, etc.

The papers cover the theory of combining outage statistics with contingencies in order to cacluate reliability indices. Also, they cover the theory of using monte carlo to calculate those indices in the absence of outage data.

But ... big but ... I still can't quite visualise how it all fits together from a practical point of view.

I tend to learn by doing so does anyone have some practical exercises that I can use to work through please?

I'd like to work through a monte carlo exercise from start to finish in order to see how to actually calculate the reliability indices. Also, I'd like to work through a similar analytical exercise (i.e. an example that uses outage data and an enumerated list of contingencies). I have access to Excel, Matlab and Digsilent powerfactory.

Thanks heaps
Andrew

 

[Skogsgurra]

Now! That is something that would also like to see! And I do not think that Andrew and I are the only two. Anyone proficient in this that could start a little seminar here? Start with a little chunk of basics and also whys. Then some whats and hows. Always with practical examples. That would be very interseting. I can think of a few guys in here that could do that. Please educate us! It will probably take a month or two. But it will be worth it.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[slavag]

Hello.
I'm with you.
It's intresting.
Regards.
Slava

 

[tinfoil]

I have a copy of this book:

Endrenyi, "Reliability Modeling in Electric Power Systems" Wiley & Sons, 1978

I was told it is considered a very good book on the topic.
Its written for a 'techie layman', and I would recommend it.

 

[Eleceng01]

I have some experience in this field, but not sure how to share it. I completed an electric relaibility study for a refinery recently using ETAP.

I can say that there are some tricks to it and the IEEE Color books were of little help (though they did make for a good start on equipment reliability numbers).

In ETAP, there are several tricks and things that can cause you issues, so just ask if you are planning on using ETAP.

Sorry I couldnt help more. I too was thrown into this when our customer asked us to do it (and of course the Principal Engineer said "No prob.")

 

[magoo2]

Richard Brown has a pretty good book providing background material on reliability planning.

I think its a lot easier if you start with a simple feeder model rather than trying to understand the whole system implications. With a single circuit, you have a variety of possible outage causes.

Each outage cause leads to an outage event, and along with that, you get a count of how many customers are interrupted and for how long. These parameters feed right into the calculation of SAIDI and SAIFI. For the circuit, you have a count of customers served.

SAIDI(hrs) = sum(customer-hours) / customers-served

SAIFI = sum(customers-interrupted)/ customers-served

These are the essential metrics that you track on a circuit level and on a system level.

From past history based on a few years, you can back into an annual outage rate per mile (or per km) that applies to a particular circuit. For example, with a 10-mile circuit and an outage rate of 0.2 faults per mile, you can expect 2 outages per year.

If the circuit has no protection other than the substation circuit breaker, then you'll expect to get 2 outages per year and these would affect all the customers. So you don't need Monte Carlo for this part.

Where you can use Monte Carlo simulation is with the circuit broken down by protective zones. Assume you have a recloser a few miles from the substation. Further assume all the taps are fused. Under this condition, you'll need to get a breakdown of customers connected in each protective zone.

For each zone, the circuit miles of exposure times the annual outage rate determines the expected number of outages. So you'll get 3 numbers:
X = # of outages in the breaker zone
Y = # of outages in the recloser zone
Z = # of outages in the fused zones

You'd actually want to do the fused zones individually because the customer counts would be different, but conceptually the 3 zones would give you a feel for what goes on.

Based on the mileage exposure of the three zones, assume X represents 10 % of the faults; Y represents 40% of the faults and Z is 50 %. So the Monte Carlo approach sets up a table of values such that 10 % of the time you have a breaker zone outage, etc. The major dynamics here is that the fused taps result in different numbers of customers. Also you can change the restoration time for each device based on history. For example, fused taps in our area require more patrolling effort before re-fusing. A tap fault would be restored much later than a mainline fault.

Run this for 100 cases or more and see what SAIDI and SAIFI result.

The alternative is to look at an analytical approach where you have outage data. You'll have outage date, customers interrupted, duration of outage, cause and number of customers served. If you take a one-year look at the service performance, you likely won't match the Monte Carlo results! You'll run into outage causes that show quite a bit of variation from year to year, so my suggestion is to pick a 3-year average look. Tree outages are impacted by our tree trimming cycle which is 3 years and lightning varies from year to year and a 3-year average also works well for that.

This is very simplistic but I think it gives you a sense of what's involved with a reliability assessment.

From a system perspective, give yourself of few hundred of these circuits, each somewhat different, with tie switching between certain circuits, with more reclosers than 1 per circuit, with chances for devices not performing properly and with the ability to change parts of the system, you'll see why the Monte Carlo approach is used.

To properly understand what is typically driving the Monte Carlo analysis, I find it easier to go back and analyze a few of the worst performing circuits in detail.

 

[catserveng]

You may or may not find this source helpful, but has been used by several of our customers recently with good results.

http://www.usace.army.mil/publications/armytm/tm5-698-1/toc.htm

Click on this link and there is a menu, you can download the whole publication in .pdf format

 

[slavag]

Hi Catserveng.
It's very helpful!!!!!
Regards.
Slava

 

[andrewward]

Hi All,

this is all great stuff. Thanks for this. 1-2 days ago I found this engineering statistics web-book. Very good.

http://www.itl.nist.gov/div898/handbook/index.htm

However, I still haven't found an actual matlab/mathcad/Excel example that I can have a look through. I'm just not sure how to set up a model. From a practical point of view, I've read a few papers and now know that I need inputs such as probability distributions, a monte carlo algorithm and some sort of representation of the power system. But, from a practical perspective I don't know how to combine them into an actual working model. I was hoping to look through someone's code or workbook or case study (in Matlab, Mathcad, or Excel or Digsilent power factory) so I can say "Oh, is that how you do it. Now I understand! "

Magoo2, thanks for the info from Richard Brown. I attended the IEEE conference in Tampa in June. He presented a paper in a tutorial that I attended. His paper seems to be a good summary of the book you refer to. There were some good comments on when to use analytic method (a specific list of outages) versus monte carlo method:

"Monte Carlo Simulation is similar to analytical simulation, but models random contingencies rather than expected contingencies. This allows component parameters to be modeled with probability distrbution functions rather than expected values. Monte Carlo simulation can model complex system behaviour, non-exclsive events and produces a distrbution of possible results rather than expected values. Disadvantanges include computational intensity and imprecision (multiple analyses on the same system will produce slightly different answers). In addition, Monte Carlo simulation is not enumerative and may overlook rare but important events."

All good stuff, but I can't visualise how to get started in converting these formulae into a working model. I also need to have a good think about how to apply all of Brown's work to the transmission system. Brown's work is for the distribution system, which has many customers per substation, whereas our customers are the distribution companies so we typically only have one major customer per substation, with a couple of other industrial or generation customers at some substations.

Great conversation though. Lets keep going

 

[jmbradley]

Andrew

Attached is a combined Markov Chain/Monte Carlo simulation mathcad11 sheet for a 19 step process that I modeled a few years ago.

It was a semi personal project, so it is not well documented.

Joe

 

[andrewward]

Thanks for the help Joe. I'll have a good look through your simulation.

Best regards,
Andrew

 

[andrewward]

Hi All,

I thought you might like to see where I got to on Probabilistic Planning. I had to write a report for my company and for a New Zealand industry body (

http://www.eea.co.nz)

on my attendance at the IEEE PES conference in Tampa. So I spent my energies on understanding the process of Probabilistic Planning and how it impacted on my role as a Transmission Planning Engineer in NZ.

I had very limited understanding of Probabilistic Planning before I attended the conference. So the report summarises what I learnt. I hope it is helpful.

Kind regards,
Andrew

 

[jghrist]

Excellent report, Andrew. A star for sharing it with us.