Expert Systems used to
disseminate complex
information in agriculture and horticulture
By Peter Gillard, Department of Primary Industry and Fisheries
of Tasmania. This story was also published in PC AI magazine.
The Department of Primary
Industry and Fisheries of Tasmania has used Expert Systems technology
to assist the delivery of information to farmers since 1992. Farming
in the developed world has become technologically advanced, and,
as a result, the decisions that farmers face are more complex. Printed
notes rapidly become out of date and for these reasons we looked
to a technology solution.
An early application was the development of "Pasturepak" an expert
system designed to enable counter staff at seed retailers deliver
the Department's expert advice on pasture seed mixtures and appropriate
cultivars (varieties). The staff get questions from farmers who
wish to sow pastures for different purposes, climate and soil types.
Pasturepak was developed using a DOS based Expert System shell called
Crystal. An advantage was that the Department did not have to rely
on the farmers owning computers. (Peter Gillard 1993 An Expert System
to advise Tasmanian farmers on pasture mixtures and fodder crops
Proceedings of the XVII International Grassland Conference, New
Zealand 763 - 4).
By 1997 we wanted to migrate to the Windows operating system,
and searched for a product that would suit our need. We selected
XpertRule® because it was a true Expert System, rather
than a front end for a data base, and it would run on Windows 3.11
as well as Windows 95. It also had connectivity with MS Office programs,
plus ODBC, DDE and OLE. The software looked truly professional with
the ability to include Winhelp, and the appearance of the runtime
products on screen was better than others that we considered. Although
XpertRule was not the cheapest Expert System shell on the market,
our experience with Crystal had taught us that a good shell saves
in development time and would be the cheapest option in the long
run.
It was not possible to read the rules from Crystal into XpertRule.
This did not deter us because the real work in developing an Expert
System is the time spent in developing rules and the models contained
within an Expert System. It took us a short while to learn XpertRule,
because it is a new technology with a different structure. In XpertRule
the logic is broken down into "Tasks" and these may be forward chained
or backward chained, concepts familiar from previous development
with Crystal. We found the graphic representation in XpertRule brilliant,
and the ability to include code has not been lost. In XpertRule
the code is bundled into "Procedures" and it is easy to keep track
of where these are in the program. Building dialog boxes is quick,
and there is a facility to duplicate and edit for even more efficient
development.
The "Apple Thinning Program"
Apple orchards are important in primary production in Tasmania,
and there has been a long history in the process of apple thinning.
Apples are naturally biennial bearing,. Trees flower heavily one
year producing a large crop of small fruit (called the "On" year)
followed by light flowering the next year with a small crop of large
poor quality fruit.
Thinning is most economically done by applying sprays of chemicals
that act similarly to plant hormones and cause the abortion of flowers
and fruitlets at an early stage of development. Early thinning favours
the development of the desirable high density of cells in the fruit.
Orchardists have to make the difficult decision about the concentration
of the thinning agent at blossom time. If the concentration is too
low, then thinning will not be effective and the cost of hand thinning
is prohibitive,. And if the concentration is too high, then there
is the risk of loosing all of the fruit. The decision is made more
difficult because of the large number of variables that need to
be taken into account.
The Department decided to use Expert System technology to encode
rules about how to make the decision on apple thinning. The research
has been published in scientific journals, and this, in conjunction
with other related research in Europe and the USA, was our basis
upon which to build an expert system. But, as so often happens,
the published data was insufficient to make specific recommendations.
The researchers, who were also giving advice to individuals and
groups of growers, were classic examples of experts who were good
at articulating their expert knowledge, but less clear at explaining
how they arrived at the advice. (cf XpertRule "Structured
Decision Tasks Methodology").
In the knowledge acquisition sessions we developed models from
tables on the white board. These are not the same as the "Truth
Tables" in XpertRule, but numerical tables that we use to capture
the range of response that the experts expect from one or other
of the variables. In developing these tables I encouraged the experts
to look for patterns. In this way we assigned numerical values for
the variables important in determining the concentration of the
thinning agent. These variables are:
- About the trees themselves - cultivar, rootstock and age.
- About the physiology of the trees - previous crop, vigour, number
of blossom buds.
- About pruning - severity of detailed pruning, limb thinning,
and penetration of light into the canopy.
- About the market - size of fruit required for the market.
- About spraying - type of spray machinery and volume of water
to be used in the machinery.
The Apple Thinning Program gives advice on five cultivars with options
that encompass four chemical thinners. The application in XpertRule
has 60 tasks, (some with 50 decision tree leafs (i.e. rule paths),
plus 30 other variables and 40 procedures supported by a customized
help file of 5,000 words. The program has made a significant contribution
to the way apple growers who have acquired it can access advice for
their spray thinning. The knowledge engineering of the problem itself
has changed the way in which those involved in spray thinning research
think about new experiments for other cultivars and possibly new spray
thinners.
In its first year there were some 20 Tasmanian apple growers who
have acquired the Apple Thinning Program, but next year this will
increase as the benefits become more widely known. The Department
has the view that its Expert Systems are "living software". This
means that there is regular revision of the knowledge. In the case
of the Apple Thinning Program, knowledge of thinning emerging cultivars
is added. Because of this the license period for the program is
limited to one or two years.
XpertRule has proved to be a very good platform for delivering
this advice. We have found that, in addition to the obvious advantages
of having a Windows environment, our program is far more robust
than it was in the DOS environment with the Crystal development
tool. The visual way in which the knowledge is laid out in XpertRule
is excellent. In our opinion XpertRule lives up to its claim of
being a tool that is able to maintain knowledge as well as deliver
it.
Read how another expert system by the Tasmanian Government won
a first prize in the Agricultural Software Competition at the Royal
Easter Show, Sydney in March 1999.
|
