The above is mentioned from time to time in various posts (usually by Railway Man). So for those of us who’d like to know more, here’s the link to the application’s vendor - Berkely Simulation Software - and a brief description of it as applied to an actual case study from 2002 that I ran across:
http://www.berkeleysimulation.com - one of those graphics-intensive with sound intro pages (usually requires a “Flash” player or similar, and takes a long time to download [|(]);
And then in the left margin or border there are 3 other web page links that are pretty interesting.
The “case study” is the “Appendix I - Rail Operations Analysis Methodology” background discussion from the July 2002 Northwestern Pacific Railroad Financial Feasibility Study - "Draft Final"by Parsons Brinckerhoff, at:
Oh, by the way - this may be as close to the RTC model as you actually get, unless you’re either “in the business” or independently wealthy - the software license fees - per desktop/ laptop" - are a "mere $45,000 !
Takes a rather expensive laptop or desktop to run it, too. Unless you want to wait a week for an answer and hope the computer doesn’t crash in the meantime.
Paul:
Thanks for the link. Boy, that sure was fun. Good thing they dont have a “lite” version of that as I wouldnt get much work done.
RWM, is that used as a simulator for projections and “what ifs” or is it used in conjunction with railroad operations? Which brings me to another question…how are dispatchers monitored and graded for their work? Are there simulations run (daily or as needed) with supervisors grading or critiquing the dispatcher?
Which brings up another question. At what point will a dispatcher simply line up the trains and monitor the artificial intelligence handle the movement of trains over a district??? or is that how it works now.
The visuals on the website were interesting, particularly the Chicago Rail Improvement Study with actual rail lines in Chicago.
Back in the day, it was always interesting to spend time in towers where you really captured the feel of a railroad. It was much more interesting than being trackside with a camera. Of course the ideal was being in a tower with a camera.
Do you ever see the time this kind of information would be available for the casually interested person? I doubt it because of security issues, but look at the availability of information now vs 10, 20 and 30 years ago.
So who generally owns or licenses the RTC software? Is it consulting/engineering firms or the actual railroads? My guess is the consultants/engineers.
Does the purchase price include the data or nodes for each line, or is that individually loaded for a segment of line? What about the actual testing of capacity issues? Are actual events used (such as real trains with actual hp/tonnage/length, etc) or are these hypothesized?
RWM, why determine the optimal times for unconstrained trains? What is that purpose? Is it for determining a baseline of costs for operations?
What are “adverse RTC models”?
Finally, proprietary costing issues…are those more dangerous in the hands of competitors or customers? My guess is the railroads all know the costs involved for operating not only their segments, but the competitors.
When you determine the optimal times for unconstrained trains, you are essentially finding the upper bound on a performance measure, i.e., that’s the best you can do. This is a very important piece of information when putting together the bigger picture, knowing that what you probably will do is going to be worse than that.
The Class 1s. A few of the big commuters. A couple of consulting firms that have a rail operations specialty practice. A couple of independent experts. The STB. The FRA. A couple of research arms of the industry, federal government, etc.
No. That’s 100% user input. It’s expensive to do it – $1,000 a mile is a ballpark number. Once you do it for a line you save it, because you do not want to do it twice.
Both historic actuals and proposed future, and blends of both.
Chuck answered that nicely. It’s provides a base line for delay not cost.
One you do of someone else’s railway for rate cases. Or other disputes.
I was wondering about the cost of inputting the data with respect to the software - so for more than 45 miles the cost of inputting data exceeds the cost of one seat of the software - which doesn’t really surprise me.
$1000/per mile. That is certainly expense data, but as someone once said “knowledge is good”.
How detailed is the input data? Certainly it would include all grade crossings, bridges, etc, but does it include MOW information such as weight and age of rail, the last resurfacing, etc?
I can see where RTC would be valuable for analyzing and potentially eliminating bottlenecks by eliminating restricting speeds, adding or lengthening sidings, adding crossovers (upgrading crossovers), adding additional revenue to a line, etc. Hard decisions can be made on capex based on ROI.
How does it account for terminal activities and delays? Is that all input data?
Is it even used for ROW improvement windows? Decisions on the types of motive power to be leased/purchased?
How are outlier events handled? Are Monte Carlo simulations used?
No. Not important, only the effect (ie slow orders or track speed) would be entered.
Terminal delays can be input and the entrance time varied for each train.
Yes.
Generally no, but it will tell you how changes in the HP assigned to a train affect the operation.
RTC is very slow to run (there is a LOT of things going on), it runs one scenario that the user has set up, it runs exactly the scenario the user has set up, optimizing the set of speed restrictions, physical constraints, train mix, train sizes, entry times and scheduled delays the user has entered.
You set up the base case (normally present case), run the simulation. Then set up a case with the changes you want to measure (new track, increased HP, new signals, different train size, etc) and run the simulation again. Compare the results.
There are additional software packages to measure/simulate train handling and fuel consumption and for network car flow.
Yesterday (04-07-2009) over on the “Sunset Route Two-Tracking Updates” thread there was a little analysis of the UP’s ex-SP route from Colton, Calif. to El Paso, Texas (approx. 792 miles) in terms of the amount of 1 main track (292 miles, 38.3 %) vs. 2 main tracks (470 miles, 61.7 %), based on a list of 18 segments - ranging from 1 to 275 miles in length - which apparently change from 1 to 2 main tracks and back to the other again approx. 15 times over that distance. For the details, see Page 14 of 15 (currently) of that thread, at:
This morning (04-08-2009) I posed the following 2 comments there, as part of the discussion with john edwards:
" Now, onto a more interesting (to me, anyway) aspect of this, as follows:
Suppose I’m a BNSF executive looking at this publicly available data, and thinking about how it affects the competitive positions of BNSF with UP (ex-SP) on that route, roughly from Texas to LA, or that portion of a longer TransCon route. What I realize is that UP has to add about 292 miles of 2nd track to be entirely 2 main tracks, including at the Colorado River crossing. At $1 million per mile - which may be low since most of the easy parts are probably done already - that’s a $300+ million capital expenditure for that alone. That’s a significiant chunk of UP’s annual CapEx budget - they could probably do it in a single year, but only by deferring other more routine items or adding that amount to the budget. More importantly, I think it’s hard - for a variety of reasons - for most railroads to
Simulation and reality are not the same thing. It’s a powerful tool. Tools have limitations. It’s not reality, it’s just a way of measuring reality. An appropriate analogy is the road maps distributed (back in the good old days) by the oil companies. It will help you get to where you need to go, but it won’t drive the car for you.
What I keep having to explain to people who have never been around railway operations is that RTC is only one possible scenario in a world of billions of possible scenarios. It uses a set of “rules” for determining meet-pass events for trains that are not remotely as sophisticated as what a human can do (nor as badly as a very poor train dispatcher), but what it does do, it does consistently.
RTC does not model anything other than what you tell it to model, and the real world is filled with variations that are unpredictible in their time of occurrence, nature, effect, and scope. If you build a model that “proves your trains run” on a given infrastructure, and actually expect the real railway to run just like that, you’ll be sorely disappointed.
I can mimic the RTC model on paper. All it consists of is a set of meet-pass events based on a set of rules for those events, on a given infrastructure, with given train characteristics.
Input whatever you want – work window, train delay, whatever. But because outlier events are by their nature unpredictible, this is fairly meaningless unless it is a regular event such as a work window.
Not in the model software itself. It’s not an appropriate vehicle. You are not trying to model reality, you are simply trying to reduce a week’s worth of paper calculations for one scenario to a 10-minute model run, so that you can compare one scenario with another scenario. This is not a magic black box tool, it’s a paperwork reduction device.
Ed, your questions are quite similar to everyone else’s about RTC. You and everyone else is hoping that someone will invent a device that predicts and measures unpredictibility down to the level of specific events – a device that would tell you exactly when to sit at the 21 table, push $100,000 in chips out there, and the next hand would guaranteed be a 21 against a dealer’s bust. But all I can tell you is how often, on average, that will happen. You’ll have to sit at the table and play a lot of losing hands to make the averages work out for you. The world is filled with events of random occurrence. There are averages for events (e.g., a big meteorite hits Siberia once every 100,000 years and flattens 100,000 acres), but the exact timing of when that event will happen you cannot predict without knowing the position, speed, energy, and condition of every atom in the universe, and then know what every whim every human and every animal is going to have, years in advance. If you want to model any event in RTC such as the effects of a grade-crossing collision, you can do that. But
Thanks for the very detailed explanations. The overview of the system seems pretty sound, but has limitations…as does any type of modeling, forecasting, or budgeting.
Obviously the extreme situations are hard to plan for, or budget for. That is for Warren Buffett and other risk managers (AIG excluded). By “outliers” I meant more in the line of heavy freight tonnage, extra movements, etc. I didnt word it well.
Rare is any business operation that runs like clockwork according to plan. Extreme economic conditions (both busy and slow) are tough to plan for…and dont we know that now!
Anyway, it sounds like a great tool and the explanation offers insite into the planning world. It sure makes the EJE/CN bottleneck easier to understand, at least as to how it was developed.
I see your point on the Sunset/Transcon routings. Your scenario has probably been worked and reworked several times, by both UP and BNSF. It all goes back to what RWM and others have stated about the best geographical route…I need to pickup Vance’s book sometime.
But, has it been determined that the BNSF routing is the better of the two?
And, which would you rather have…the high priority UPS fliers or the heavier stack trains? That is where the simulation could help determine your marketing mix as it applies to costs, pricing, etc.
Of course, if we have the “lost decade” that Japan experienced, it wont matter for awhile.
Is it a duplication of real events? No. The simulation world is the perfect world. Trains accelerate at their optimum rate,they brake at the optimum time, the trains are met and crossed over in a best case fashion.
Does it simulate it well enough that you can measure the difference in operation between two scenarios and make an educated decision between the two alternatives? Yes.
I’m pretty interested in computer simulation software, so I googled up
a couple articles. Footnote at the bottom will help you chase them.
This is pretty impressive stuff.
I have a remark based strictly on computer software principles. I
assert that there is a software engineer’s opportunity here.
The RTC program - at least the version I’ve read about - demands
extensive input preparation.
A paper (“Meetpass”) signed by Shughart discusses RTC and other
similar, saying of RTC “Very detailed, and complex input data
requirements requiring days of analyst effort”. $1000 per mile in
RWM’s estimation.
Is it possible that problem generation is the “rate determining” fact
that constrains analysis studies?
I suggest that the value of RTC might be much enhanced if there were a
computerized assist for this task of problem generation. This would be
no small project - would require software engineering skills as
professional as the people who built the simulator.
If it really happened that problem generation became “simple” t
Now, onto a more interesting (to me, anyway) aspect of this, as follows: