Tuesday 31 August 2010

The abuse of forecasting

Mention the name "Gene Woolsey" to operational research scientists of the 1970s and 1980s, and you will probably get the reaction that he spoke and wrote great deal of common sense, mostly in "Interfaces". One of his stories is how he did a study in a particular factory, and a couple of years later returned on a visit. He said that he wanted to creep away quietly; the solution that he had proposed had been pinned on a board, and was being followed to the last detail. In the intervening years, the environment had changed, so all the external parameters of the model were different, making the solution totally inappropriate.

I thought of Gene last week, when our gas company sent us "Your Annual Gas Statement". It reads
"We've tried to make it as easy as possible to understand."

"Your usage: From 25 Aug 2009 to 24 Aug 2010, you used 15396.49 KWh of gas.
If you continue to use energy at the same rate over the next 12 months, we forecast your cost will be £568.21"

What wonderful precision! Especially as the power consumption is based on reading a meter which is accurate to +/-1 metric unit, and one of those is between 11 and 12 KWh. So, the meter can't determine whether we used 15390 or 15400 KWh, so the last three significant figures of their record are unnecessary. That translates to making the pence in the forecast unnecessary. But these errors are tiny in comparison with the assumptions that we will continue to use energy at the same rate.

It worries me that somebody has thought that this information is intended to be useful. If it was someone from the O.R. department, then I suggest that they creep away quietly now.

If you are from, or know someone from, the O.R. department of British Gas, do let them know of the abuse of forecasts.

Monday 23 August 2010

Cutting your hedge

When I taught a course on multiple-criteria decisions, I used the frequency of hedge cutting as an example of conflicting objectives. Should you cut the farm hedges each year, or every two years, or every three years? The more often you cut, the more it costs, but each cut costs less. And hedge-cutting around a farm happens close to harvest time, so there are other annual tasks to schedule. The location of the hedge also matters. Those by tracks need more frequent trimming than others.

Visiting Old Walls Hydro reminded me of these, because the site has just achieved a high standard of conservation, and one requirement is that hedges are trimmed every three years. But there is a complication which I had overlooked in my course. In three years, some hedgerow trees get so large that a sapling in year 0 is a substantial tree in year 3. Should you trim that tree? Or allow it to grow to be a tree?

Models for hydroelectricity

Last week Tina and I visited the Old Walls Hydro site in Ponsworthy on Dartmoor. Water is taken from the West Webburn river and diverted along a leat to give a 16 metre head of water for two turbines. As a piece of engineering it is fascinating.

Various aspects of the design reminded me of some operational research principles. For parts of the design, the owners were told that what they proposed was impossible. How many times does an O.R. scientist or team meet such prejudice? Too often. Sometimes the problem is posed with a feasible region that is too small, simply because nobody has pushed the limits. Beware of incorrect definitions of feasibility!

Then, there were matters of feedback. One of these concerned filtering the water before it entered the pipes to the turbines. Leaves and other vegetation falls into the leat and is trapped on a wire mesh conveyor belt. Once sensors detect that there is a difference in the level of water before and after this screen, a motor starts and removes a "belt-load" of debris. How should this be powered? The simple answer would be to take the power from the turbines ... but that can be affected by the presence of the debris and could cause the system to collapse. So, the motor is isolated from the hydro power. It runs on a small battery, which of course is then trickle charged from the turbines. The O.R. lesson is that feedback needs to be controlled, so that it gives useful control at all times, and not in ideal conditions.

Thursday 12 August 2010

Rubik's cube

Rubik's cube is not mainstream O.R., but news about it has made the news today, with the proof that every scrambled position can be solved in 20 moves or less. The announcement is here. It is good when mathematics is covered in the press.

There is a slight link with O.R., but I won't exaggerate it. In theory, you could create a dynamic programming formulation of the problem of solving a scrambled cube. The state would be the description of the cube; the decision would be which of the possible moves to make next; the single stage cost would be one; and the objective would be the minimum number of moves needed to solve the cube from its present state. Stages would be identified with the objective value. But before anyone tries this, note that the researchers looked at 55,882,296 cases (states).

And for the last two years, I have shared an office with Peter Vamos, one of Erno Rubik's cousins.

Wednesday 11 August 2010

Farewell to all that

Today I removed my last possessions from the university campus. Since I retired, I have had the privilege of a little working space on campus, in an office shared with several other active retired staff. I have used it for writing and editorial work, and valued the chance to see and talk with former colleagues.

Alas, that has now come to an end. The office is no longer to be free for the honorary staff, and so I decided to move my last few things home. As I did so, I thought back over the 35 years since I came to Exeter. It is unusual to have spent so much of one's entire working life in one place.

I have had six different offices, in three different buildings.

When I came, the data and programs for my research came with me in two card cabinets, each holding 10000 punched cards. When I left, I needed a USB stick to carry far more data and material. I regret that one of the datasets that was on punched cards has been lost; it can never be replaced, and there was scope for some more statistical analysis to be done on it. Some of the research publications that I had used for my PhD research were on microfiche; that is a medium which has practically disappeared. When I came, I already had a small collection of books and journals that filled about 12ft of shelf space. Even with pruning, I have three times that space for professional material.

When I started, the language of choice for programming was Fortran. That was my fifth language (after Algol 60, Titan Autocode, Basic, CSL). We progressed through Simula, Pascal, C and C++. The first desktop computer we used was a PET, but we never did any research on it. A research grant provided a more substantial machine in 1981 or 82, which used 8inch floppy discs, and compiled a Pascal program in about 20 minutes! Olivetti PCs arrived in 1986, and thereafter most of my computing was based on PCs.

What changes will there be in the next 35 years?