Slashdot Mirror
Genetic Algorithms Improve Combustion Engines
University of Wisconsin Madison's Peter Senecal has
evolved a new
combustion engine which cuts nitric oxide emissions three-fold, soot
emissions by fifty percent and fuel consumption by fifteen percent. His genetic
algorithm searches for the best combination of six parameters which determine
the design of an engine. It starts from a search space of five, and includes
strong heuristics to minimize the search space considered.
Listen, a GA doesn't care about hills in the search space. This isn't a gradient based technique (hill climber). The strength of a GA is in it's ability to search a large design space and not get stuck on local maxima.
-Put a stop to procrastination... Later....
My main concern is whether the engine can work well in the real world. Getting efficiencies in a test cell is easy (I have an acquanitence who designs engines. He has a test engine with variable compression ratio. On a good day, he can keep it running for a while at a compression ratio of 30. It's a *long* way from being roadworthy.) The variation you see on the road is much larger than in a test cell, or in other applications, like aerospace--conditions that a car engine take for granted could raise hob with an airplane engine. F'rinstance, wildly changing loads over a matter of a few seconds.
Still, the GA design methodology sounds interesting. I wasn't clear how they avoided getting stuck on local minima. Is this what the 'mutation' handled?
I don't recall seeing crash test results for the Lupo in particular, but I've definitely seen them for other small cars. They're by no means universally worse than larger cars - in some examples they're better. Bigger cars can be safer but they're by no means guaranteed to.
Greg
(Inside a nuclear plant)
Aaaarrrggh! Run! The canary has mutated!
To Americans this may seem strange, but many small European cars do 70-80mpg. A friend of mine has a Volgswagen Lupo (small, but five seats) that he drives 100 miles a day. It does 85mpg.
The Lupo is actually not street legal (I'm pretty sure it was the Lupo) in the US.
It's too small or too light one, I can't remember. But I wanted to have one imported for the obscene fuel efficiency. Instead we got a diesel new beetle and get about 50mpg.
But remember, small doesn't mean fuel efficient, my Geo Metro is only a little bit bigger than the Lupo and only gets about 30mpg. And of course things like the Z3 and the Miata get about 15-20mpg.
Kintanon
Check out JoshJitsu.info for Brazilian Ji
Try Ascend from CMU.
http://www.cs.cmu.edu/~ascend
They have a very good (although very steep learning curve) nonlinear & differential equation optimizer. Handles thousands of variables.
I recall that Danny Hillis (ex of Thinking Machines Corp, now works at Disney, I believe) used to do GA stuff. He and some buds made a GA that generated sorting algorithms. But it would get stuck on local minima, and would settle on suboptimal solutions.
So they added another element to the environment -- another set of GA-bred algorithms that generated sets of numbers to be sorted. Their goal was to create data sets that made the array sorters perform poorly! Excellent!
So whenever the sort critters found a nice local minima, the nasty data set generators would find their achilles heel and chase them all away from that area.
I really liked the predator/prey flavour of the idea.
Regards, your friendly neighbourhood cranq
Regards, your friendly neighbourhood cranq
About 25 years ago there was an experiment done by one of our fellow programmers who decided that computer software can be programmed by using Darwin theory of evolution. A large mainframe was programmed to create small pieces of machine code (functions) and then the generated code would run a series of tests and the tests were designed to grade the selected code by utility, the heuristic was accomplishing sorting of a character string. Basically, a generated function would run on input provided by the tester code (where input is a randomized character string) and the output would be compared to a sorted character string. If a strain of code produced output that looked a little like a sorted string, this function would be stored for the future generations. After running this tests for a week, the best family of generated functions was fed back into the mainframe, the test functions were adjusted and some strains of code were introduced into the machine that were written by the programmer that would help to speed up the process. The functions were allowed to mutate and to reproduce by combining features from some functions into new ones. At the end a computer program was generated that sorted the entire string of characters. Interestly enough, the programmer could not figure out exactly how the string was sorted, the software was just too complex to understand (I supposed he did not want to waste time trying).
I think anything at all can be accomplished by GE, the only drawback would be that us - humans - may fall out of the production loop. We will not have to understand why an engine with octagonal and hexagonal and other types of parts work better than something else. It will be too complex for us to understand, and even if we could, who would bother? We would just use the results that appear as if by magic.
But I guess there are drawbacks in all methods...
You can't handle the truth.
Wonderful. Good for you. By all means, live 20 miles from work and burn a couple of gallons of non-renewable, environment-destroying fuel on the way to and from work every day. But, PAY FOR IT AT A REASONABLE RATE. Hey, I pay $1,600/mo for a 400 sq ft apartment in Manhattan, and I'm getting off lucky in this market. I don't complain.
The fact that SUVs can exist, and that so many people can drive them, means that gasoline is simply too cheap, when you take into consideration the damage it does to our environment. Not to mention the damage in quality of life and general integrity of our nation that is done when everyone just gets fatter and fatter and lazier and lazier due to the ridiculously low cost of gasoline.
According to the article, Caterpillar needs a solution that cuts nitric oxide emissions in half by 2002. So you may see this innovation sooner than you would expect.
Your right to not believe: Americans United for Separation of Church and
-Put a stop to procrastination... Later....
see subject
If it ain't broke, fix it 'til it is!
These are my friends, See how they glisten. See this one shine, how he smiles in the light.
The dimples on golf balls act to trip the boundary layer from laminar to turbulent. This transition pushes the point of boundary layer separation to the trailing side of the ball, which reduces the pressure drag on the ball. Note: the transition from laminar to turbulent flow is probably the result of the dimples exciting instabilities in the BL
Now, as for flow control techniques for drag reduction and separation control, this is currently the most active area of fluid dynamics research. I just returned from the AIAA meeting in Denver and participated in a poster session where Boeing/Pratt & Whitney were showing off their C-17 project. Using pulsed jets of air to mix out the jet shear layer more rapidly. United Technologies Research Center had a poster, as well as a nice, talk discussing dynamic flow separation control on helicopter rotor blades. A group from UCDavis was showing off some MEMS MicroFlaps that they are investigating as potential replacments for the large overweight and extremely complex high lift devices found on most large aircraft. Then there was a group at Notre Dame, investigating Phased Plasma actuators (think surface mounted glow plugs) as a means of controlling high speed flows.
The passive techniques used in golf balls is also a large area of interest. I have spent a while talking to some folks from Princeton who've been making measurments for Callaway, and attempting to improve the flight characteristic of their balls. Someone mentioned vortex generators, and feel compelled to mention that NASA Langley, as well as a number of universities, has been playing with MicroVGs for quite a while now. There is also a group of folks using dynamic VGs and fluidic VGs (typically referred to as Synthetic Jets) in the research community.
The biggest problems with the application of flow control techniques to practical problems are size, weight, power requirements, and robustness. I could probably reduce the drag on my Honda appreciably within a couple weeks, using some of the stuff I play with in my lab. BUT the overall fuel efficiency may not make improve noticably due to the power requirements of active controll and passive techniques are typically tuned to a specific set of flow conditions. Then there's the car wash issue...
But what about the cost of mining the ore? Surely it must take quite a bit of energy (most likely in the form of, once again, gasoline) to run the machines that mine the ore. And that transport it to the place where it is smelted. And that transport the resulting steel to the factory.
Of course, the people who do all of this have to get themselves to work, which means more gas burned.
And then there are all of the plastics, and electronic equipment that go into cars. Not only is there a cost in terms of the chemicals and energies needed to produce this stuff, there's the cost of disposing of this stuff when the car is no longer needed.
Finally, the resulting car has to be shipped typically several thousand miles (at least) to the dealer. Surely there is quite a bit of fuel being used up in this process as well.
From this vantage point it really looks to me like burning gas is actually more environmentally friendly than building the car which is going to burn it.
Genetic Programming and Genetic Algorithms really work. Here's the general idea behind genetic algorithms (and a specific example - curve fitting)
1)Express the problem and solution space in terms of a set of numbers.
ex: coefficients on x^i where i steps from 0 to 100.
2)Express a fitness function - this can be very difficult!
ex: testing 1000 different points, fitness = sum of standard deviations
3)Generate a random set of hypothetical solutions to the problem - it's best to generate 100-1000.
4) Test the fitness of each possible solution.
ex. just as stated in 2, sum the standard deviations.
5) Keep all the solutions so far (within reason) and add:
5a)Some mutations of some of them.
ex. change some of the coefficients a bit.
5b)Some crossovers of some of them.
ex. take some coefficients from solution X, and THE OTHER COEFFICIENTS from solution Y.
Note: mutation and crossover policies have to be well designed so as to stop local minimum issues.
6)Go back to 4) until the fitness of a solution is within some threshold of the ideal fitness
(in my example, that might be 10.000000 or something).
Check out the following resource for source code if you want to try it out yourself:
http://www.aic.nrl.navy.mil/galist/src/
There are FAR more "everyday joes" destroying the environment than rich people.
I think that when a fine is used as a punishment (for example, with speeding tickets), then the fine should be based on income, because it is the only way to make the punishment as effective for everyone.
But I'm not talking about a fine - I'm talking about everyone paying the "true" cost of gas, per gallon, when taking in consideration the damage being done to the environment as a result.
As an antenna engineer I use GA's all the time to optimize antennas, filters, polarizers, transitions and other microwave related stuff. The reason it works so good for these problems is that the search space is huge and filled with local minima. I have found that it's always a good thing to use a simpler algorithm such as a hill climber or some sort of gradient based method to really squezze the last few tenths out of the cost funtion though.
GA's are great in finding areas of interest but converge very slowly. Especially considering that electromagnetic simulation is very expensive in terms of memory and CPU.
Of course we run all our optimizations on a Beowulf!
http://www.endwave.com
I had originally used a table of the 15 most common English trigrams (available here), which was not giving me precise enough scores. Then a friend I had met through our mutual struggles to solve the stage sent me his trigraph table. In his words, this is how he described it:
I used Project Gutenberg and downloaded the complete textversions of Bram Stoker's "Dracula" (800KB) and "Wuthering heights" (600KB). I created a huge string out of the text (eliminating everything which is not a - z or A - Z) and ran a window of 3 chars over it, each time noting how many times a particular trigram appreared. I mapped that in an array. Doing like that I created a textfile of 26^3 trigrams and their respective scores (log2(N+1) where N=number a particular TriGram appeared in my sample text. (e.g. The score for 'THE' is about 14.0.)
--
Have fun: Join D.N.A. (National Dyslexics Association)
I've been wanting to do something like this for years...too bad someone beat me to it. (well, I'm sure there have been other, similar things done as well)
[pink beam of light]
Wonder how much the oil cartels will pay for this one.
Finkployd
--
Ancient Goth: Someone who overthrew the Roman Empire.
Time is Nature's way of keeping everything from happening at once... the bitch.
Actually, it's because us Europeans use bigger gallons
1 US gallon = approx 0.83 imperial gallons
- Andy R
A pizza of radius z and thickness a has a volume of pi z z a
But what about the cost of mining the ore? Surely it must take quite a bit of energy (most likely in the form of, once again, gasoline) to run the machines that mine the ore. And that transport it to the place where it is smelted. And that transport the resulting steel to the factory.
This does indeed take energy; however, the point to bear in mind is that smelting takes a *huge* amount of energy - comparable to the chemical binding energy of the ore (for obvious reasons).
One kg of ore has a chemical binding energy in the realm of 2 MJ (assuming 200 kJ per mol of oxygen molecules stripped).
By contrast, to haul that 1 kg of ore and 99 more kg of rock bearing it up a 1 km mine shaft takes about 1 MJ. And that's under pretty extreme conditions.
And then there are all of the plastics, and electronic equipment that go into cars. Not only is there a cost in terms of the chemicals and energies needed to produce this stuff, there's the cost of disposing of this stuff when the car is no longer needed.
Producing plastics is cheap - it's just fractional distillation and catalyzed reactions, neither of which take up much energy.
Similarly, disposal is cheap, as there isn't much hazardous waste in a car (just the battery, mainly).
Again, the important thing to bear in mind is how mind-bogglingly dense chemical energy storage is. That's why smelting is so substantial a chunk of the energy cost of building a car, and that's also why even the smelting cost is dwarfed by the gasoline consumed in driving the car.
Finally, the resulting car has to be shipped typically several thousand miles (at least) to the dealer. Surely there is quite a bit of fuel being used up in this process as well.
Not at all. Hauling a car in a transport cart is actually less energy-expensive than driving it the same distance (that's why transport carts are used). Over its lifetime, the car will have easily a hundred times that distance put on it - the dealer transport distance is insignificant by comparison.
From this vantage point it really looks to me like burning gas is actually more environmentally friendly than building the car which is going to burn it.
I'm afraid that I still disagree, for the reasons stated above. However, I do compliment you on a very well though-out argument (I don't see that very often).
Do you see? We don't go over the entire sample space at all: we take a guess, look at the area of sample space around the guess and head in the best looking direction. Keep going (with a little randomness thrown in to make sure we don't get stuck on a solution that's only better than a tiny area of sample space just around it) and we tend to end up in a damn good place. Do it several times ('cos you might just end up in a different damn-good-place the next time around) and you're left with a bunch of really good approximations to a solution. Pick the best of these. You end up having only actually done the calculations for a very few engines (sample points); you tend to have ignored vast tracts of hideously misbegotten engines that, eg, pump in 14 gallons of fuel a minute and never get hot enough to light it, that your infinite number of monkeys would have built at some stage.
nal 11
Actually, they were looking at diesel engines, not gas ones. There are a fair number of diesel applications where the lifetime fuel costs are larger than the entire vehicle cost, much less the cost of the engine alone. A big rig may pile up 1 million miles, and at 5-7 mpg, that adds up to a lot of diesel fuel.
There's no point in questioning authority if you aren't going to listen to the answers.
This is due in no small part to the constraints of golf ball design (spherically symmetrical, to name the bigest). On larger objects and/or those which can actually be optimised for one direction of movement, laminar flow is usually better for cutting drag. On the other hand, it's possible that controlled turbulent flow might improve the drag figure of a car somewhat and at less cost than other means. Then you get into little details like the technical ability to produce a nice finish on a bumpy surface, customer acceptance... the best drag-reducing trick in the world won't save a drop of gas if nobody will buy a vehicle that uses it.
--
Ancient Goth: Someone who overthrew the Roman Empire.
Time is Nature's way of keeping everything from happening at once... the bitch.
I wonder which does more damage to the environment - burning up more gas in an old car, or building a new one.
That's simple enough to estimate.
Most of the effort (not cost) that goes into making a car goes into smelting the metals used in its construction. Even very complex manufacturing processes take much less energy, and hence cause much less pollution.
The amount of energy needed to smelt the metals in a car is an inefficiency factor times the weight of the oxides you'd get by burning that metal. Lumping all of this together, it Fermis to around a factor of 10.
Let's say you have a tonne of metal (overestimate), and about 33 kg of gasoline in the tank (1/3 of 100).
1000 * 10 / 33 = 300.
If, over the course of the lifetime of the car, you fill the gas tank 300 times or more, you've caused more pollution by burning gasoline than was caused by the fossil fuels burned to smelt metal and produce electricity to manufacture the car.
Genetic algorithms can be used to optimize all sorts of problems.
For example this page describes optimization of wind turbines with genetic algorithms.
Like all engineering problems, the biggest challenge with these sorts of problems is determining the formulae to predict performance. A great deal of knowlege about engines needs to be used to develop these simulations. If you can't model what effects changes in the shape of turbines or cylinders will have on performance, then you can't build a fitness function. The fitness function is used to determine which gene sequences will "live" and which ones will "die".
If tits were wings it'd be flying around.
-Put a stop to procrastination... Later....
Caterpillar was talking about a highly advanced diesel which would break the 50% thermal efficiency figure using insulated pistons and cylinder heads, an insulated exhaust system, a turbocharger operating at 70% efficiency and turbocompounding. I heard nothing since, and have no idea what happened to it; maybe the high combustion-chamber temperatures would have created too much NOx, and the Clean Air Act consigned it to the junkyard. If so, perhaps genetic algorithms can salvage the technology and bring us some benefits (and relief from OPEC price gouging) in the bargain.
--
Ancient Goth: Someone who overthrew the Roman Empire.
Time is Nature's way of keeping everything from happening at once... the bitch.
This is true, but if you're driving an old car instead of buying a new one, you're helping the environment also, so perhaps it balances out.
...
I wonder which does more damage to the environment - burning up more gas in an old car, or building a new one. Considering the amount of energy and effort that must go into building a new car, I would say that it might actually be more environmentally sound to drive one that is old and uses more gas, than to buy a new one.
Anyway, I have no sympathy for people who whine about gas prices. If you're going to destroy the environment, then you should pay for it. And you should pay for it at a rate far greater than the rate at which you currently pay for gas in the U.S. Like, say, $5.00/gal. I would be SO happy if gas went up to $5.00/gal (as long as it wasn't just the oil companies getting rich, but instead a tax which go to something useful).
And no, I don't own a car, or any motor vehicle at all in fact (I live in NYC where they are less than useless), but even if I did, I would still want to pay $5.00 to be reminded every time I went to the pump what damage I was doing to the world. And of course, I want everyone else to be reminded of that as well.
Hey America - get off your fat asses, get out of your SUV's, and try *walking* or *biking* to work (or, if it's too far, then - heaven forbit - move closer to work!)
A GA certainly does care about hills in the search space --- but you're also right in that it's not just a simple hill climber. How can this be so? Well, a simple hill climber just looks around it and always heads upward. If you have a search space that has a tiny hill next to a huge one, and you start a hill climber on the slopes of the tiny one, it'll chug up to the top of the tiny hill and sit there.
Now, a GA throws in a random element as well. That's to say, the next step for a GA doesn't always have to be in the 'up' direction. So start a GA on the tiny hill, and if it's random enough the population that forms the next generation will be spread out all over the tiny hill and partially up the slope of the massive hill. Natural selection then comes into play, and the parents of the next generation are the guys and gals who are climbing the mountain. Next generation, the population will be spread even further up the slope --- and of course the ones at the top get to be the mums & dads...
Of course, you can see that if the GA isn't random enough (too low mutation rate, or not enough variance in the gene pool), the GA could quite easily get stuck on the little hill. This is why when we solve problems with GAs, we tend to use lots of different starting points: we know that each starting point will probably lead us to a different (but large) local maximum, so we try to get them all.
(You could try increasing the randomness. You can see where this leads: too much randomness and you might as well be doing a random search; you're destroying the 'partial solution' that your genetically-bred creatures have found at each step.)
nal 11
When using a very large, genetically diverse population it doesn't matter very much how hilly the search space is. When you are using a small population of 5, it matters a lot more. The technique described in the article is akin to taking a bunch of random steps in the seach space and following the one that turned out the best. If the terrain is too hilly, this isn't any better than randomly guessing a bunch of parameters and picking the set that worked best.
Never mind that the speed limit is enforced and the average SUV driver* has never gone offroad in his/her collective life, but we still need MORE POWER!
I don't know about anyone else, but I want power for acceleration, not top speed. I used to have a diesel MB that had unbelievably bad acceleration. It got to the point where I was literally afraid to change lanes because there was no margin for error.
I like my SUV with V8 power just fine. And yes, I want to make sure I have way more weight than you. As far as I'm concerned, it's survival of the biggest.
--
Sometimes it's best to just let stupid people be stupid.
Being in the same city as the recent World Petroleum Congress, I've been propagandised to quite a bit by the media (and hence the oil companies). Supposedly, many of the bigger oil companies are already deep into developing non-oil technologies (no, I guess they're not stupid enough to think that their oil wells are bottomless), and they're just waiting on the consumers to ask for it (apparently consumers have a nasty habit of, well, being complete hypocrites). I'd take their words with a grain of salt, though.
Or just move to a state that doesn't have smog testing to begin with. Around here if there is a place to screw a license plate onto it, it is street legal. Registration is done entirely by mail. No inspections, no smog check, no hassles.
GAs tend to be useful in discrete problems, where standard non-linear optimizers don't apply. Even there, GAs are often inferior to other stochastic algorithms. In general, the use of a genetic algorithm requires more performance evaluations than simulated annealing, and frequently more than simple stochastic hill-climbing.
There's one key exception, however. If the objective function has essentially cylindrical optima (e.g. the function f(x, y) = (1 - x^2) * (1 - y^2)), then the crossover operator allows the system to use "hyperplane search": the "crossover operator" (used in the generation of the new population members) will frequently tend to take the good parts of different candidates and glue them together, making better offspring.
What's sometimes surprising is how many objective functions can be encoded so that they have roughly cylindrical optima relative to the cross-over operator. For instance, in the old work on the Travelling Salesman Problem, van Gucht et al. used segments of circuits as crossovers, and that gives a roughly hypercylindrical objective function, thus speeding up convergence.
All this means that without actually looking at the particular objective function and encoded, we can't really tell whether the use of the GA was wise or not. It depends on the constraints of the problem.
I lately read an article about a way to cut fuel consumption to half. A little gadget mixes the fuel with water and because water will expand a lot more when getting hot, the efficiancy of the engine goes up while fuel consumption goes down. And the whole thins whould work with most modern engines. It would require just some tweaks in the injection system.
The guy who invented this hold a patent for this for years, but still no car company wants to build it....
I don't know for sure, never having talked to senecal about his research (combustion research isn't my field), but I suspect a factor might have been the amount of time it takes to test each member of the population. Each simulation of an engine takes many hours, and I believe the limit was 4 jobs per person at the time he was probably running those simulations.
It's cool to see something I was actually tangentially involved with on slashdot! ;) I used to work as the sys admin at the UW-ERC where this work was done. The "SGI supercomputer" referred to in the article is probably the Origin 2000 with 32 R12000 CPUs running at 300 Mhz and 16 gig of ram.
--Kevin, former ERC sys admin
--LeBleu
If you're reading this you're part of the mass hallucination that is Kevin the Blue.
From the very brief description in the article, the genetic algorithm seems to function very much like a simulated annealing algorithm. Can anybody comment on the differences between genetic algorithms like the one used in this case and simulated annealing algorithms? Are there any features to a design problem that would make it more amenable to one or the other?
For those unfamiliar with simulated annealing, here is a quick description: Simulated annealing algorithms need some parameters for defining a design and a function for evaluating the quality of a design with a particular set of parameters. The algorithm keeps some sense of temperature which starts high and steadily decreases through the running of the algorithm. The main loop of the algorithm perturbs the design slightly (changes the parameters) and either accepts or rolls back the change with some probability, based on the change in quality caused by the change in the design, and the current temperature.
Ben
So...the question is...who gets the patent, and what for? Can the engineer really claim that he "invented" this engine because he used GA? Should the algorithm itself be patented? Inventing machines. Food for thought.
It's 10 PM. Do you know if you're un-American?
Genetic Algorithms are a staple for engineers. About 5 years ago they used almost the same technique to achieve similar results with jet engine turbines. Civil engineers love the stuff especially. Consider the problem of finding the optimal route for a highway through mountains that involves moving the least amount of dirt. The search space is massive, but not so hilly that a GA can't function well.
-- http://thegirlorthecar.com funny dating game for guys
Important in fuel efficiency are other factors such as wind resistance, vehicle weight, and power saving devices such as efficient breaks which channel the energy created from breaking back into an electrical engine.
These fuel efficiencies are seperate to the engine, but can be co-dependant. Already cars getting 70 miles per gallon have been created simply by being dual electical/internal combustion.
As a former worker at Ford Motor Company I used a genetic algorithm to optimize fuel efficiency as a function of cost. But maybe I wasn't thourough enough... Is it possible the biggest gain is yet to come when the ENTIRE car model is fed into a genetic algorythm and optimized by geometry, with goals of fuel efficiency and vehicle cost?
-Ben
Check out the GARAGe at Michigan State University. They use Genetic algorithms for many different and interesting problems. From consumer preference predicting to financial analysis.
There code is called Gallops, and it seems very scalable. There is a Meta-GA built into Gallops that allows the GA to genetically change itself in order to be the most efficient GA.
This is cool stuff!
Keeping
One might confuse ASA with a "hill climbing" optimizer, but since it has a randomizing parameter (temperature) built in, it can be rationally adjusted to explore regions outside of local optima.
Seastead this.
I've recently learned about Genetic Algorithms (GA) in my quest to win $15,000 from The Code Book and Simon Singh's Cipher Challenge (eGroup here). One of the stages is a deft Playfair Cipher, which have historically proven difficult to solve by hand. Using a genetic algorithm, I was able to solve the cipher in just 28 generations.
What's amazing to me is that here I have just 500 lines of code that know nothing about ciphers, Playfairs and codebreaking, yet using a simple mutation and scoring function was able to break a relatively difficult cipher.
For those that don't know, a Playfair cipher puts the English alphabet into a 5x5 grid (minus 'j') and uses pairs of letters to select other letters from the grid. Instead of a 26-letter substitution cipher, codebreakers are now faced with a daunting 676 letter-pair challenge.
My code created 1,000 random keysquares and mutated them, randomly selecting squares and swapping them with one another, or swapping entire rows and columns. The new generation was scored, and those that scored high had a better chance of making it to the next generation than those that scored low (survival of the fittest, if you will). And in just 28 generations, what was once a mass of jumbled letters slowly transformed before my eyes into perfect English. Once the solution had been found I actually felt bad about killing the process, as if I had creatd life and killed it. It was truly amazing.
--
Have fun: Join D.N.A. (National Dyslexics Association)
I suspect the two will intertwine before too long. I agree that the current high prices of OPEC are a bit uncalled for (and will consequently drop), but I wouldn't get to used to it. Unless someone at OPEC starts smoking some crack and declares oil to be renewing itself, oil prices should gradually increase over the next few decades (of course, in our society, who wants a car that's going to last decades?!).
That said, an efficient engine is better than an inefficient engine, no matter what the oil prices are.
As far as I know, mutation is not the method to avoid local minima. 'Diversity', to borrow an analogous biological description, is used to prevent local minima. If you have a gene space that accurately samples the entire space, and an algorithm that doesn't kill variation too fast, you will find several local minima, statistically, without being trapped within any of them.
-AS
-AS
*Pikachu*
-Put a stop to procrastination... Later....
Rather than focusing on internal combustion engines, maybe they should apply these algorithms to other kinds of engines, like electric/solar/natural gas/etc. to see if they can come up with something smaller and more powerful. Most of these already have polution advantages.
Can anyone explain why a genetic algorithm is needed here? Conventional numerical optimizers can handle thousands of variables, even for nonlinear problems. The article says there are six parameters, not a big search space at all. Is this really an advance, or just jumping on the genetic algorithm bandwagon?
Ever hear of the Boing777 ? The jet engines were optimized by a GA. Three days worth of computations on a nice piece of hardware probably saved them three years of engineer research.
bash-2.04$
bash-2.04$yes "Don't you hate dialup connections?"| write USERNAME
Does anybody know any more about why the GA the researcher used had such a small population size? Population of 4 + 1 elite seems mighty small to me- 50 or 60 sounds more like it, even if it takes ten times longer per round to evaluate, because you typically want more diversity. Or in this problem did he find that the diversity wasn't worth it? Does anybody know?
--
-jacob
-jacob
Or rather a sad one about the current state of the Patent Office?
Cheers,
Ben
My usual seat in the cluetrain is at A HREF="http://pub4.ezboard.com/biwethey.ht
A number of scoring techniques could be:
letter frequency - the more it matches to the frequency table of the original language, the better the score.
letter group frequencies - same as above, but with groups of two letters.
the occurance X between identical characters - read the playfair cypher explanation.
a dictionary of common words in the original language.
I would really like to know more about how the original poster based his scoring system.
Okay... I'll do the stupid things first, then you shy people follow.
Okay... I'll do the stupid things first, then you shy people follow.
[Zappa]