|
Iowa | I used to back in the day. I had a five year plan with a few hundred variables and constraints that would tie up the main frame computer on campus for a while until it found the solution. Then, I did a much more simple model on a standard simplex that I wrote for an Apple IIe, and sometimes it would grind away for a day or more before finding the optimal solution. It was a fun, educational exercise in futility. The real value in it was probably more in terms of gaining greater ability to think about things in terms of a system rather than individual components. Todays computers would probably make much shorter work of it, but the reality of the optimal solution being optimal in the real world still depends making the right assumptions when building the model. A minor change in a coefficient somewhere can drastically change the whole outcome. The odds of accurately modeling the real world literally millions of variables and relationships between them is far more than we can understand, not to mention model. That's why I have a real problem putting much faith in global warming models. You can really talk a good game with them, but they are based on our simple assumptions about a complex world. That's my opinion, but I've not really worked with it much in these regards for a long time.
As for my 5 year model optimal solution, the real world markets had other ideas within the first year.
On the other hand, linear programming works well for things that can be modeled with greater certainty like balancing least cost feed rations. | |
|
Linear Programing
