A decision model for resource management using rule based utility functions and parameter selection

The joint application of decision analysis and expert systems to the problem of nutrient selection in a controlled environment enabled the construction of a decision model that accomplished what neither of these tools could have achieved alone. The ability to use a decision tree with probabilities together with intuitive modeling for the crop response function through the use of expert systems in the same model provided a flexible and adaptable system which made better decisions than would have been made by either decision analysis or expert systems employed alone.;A decision tree comprised of a decision node (which nutrients?), a chance node (probabilities of solar irradiance ranges), and an outcome (crop response function) was used as the core of the model to decide which nutrient recipe should be used at any time of the day. The expected value of each recipe was determined by computing the cross-product of the probabilities of irradiance levels and the crop response function.;Rulebases for the crop function and the nutrient selection procedure were derived by growing a cucumber crop under the advice of a plant nutritionist in a greenhouse in the winter of 1988 at the Ohio Agricultural Research and Development Center, Wooster, Ohio. Advice was obtained by means of telephone conversations and personal discussions as the plant nutritionist was not permitted into the greenhouse. Weather forecast data were assembled into a database to derive parameters for a s probability distribution function for solar irradiance derived from clear sky irradiance computations.;The model decided upon the selection and application of nutrient mixtures to a cucumber crop in a controlled environment in real time, on a repetitive basis, within the context of limited resources. It incorporated an adaptive feature to modify the parameters of the radiation rule base from one day to the next so as to reduce the variance of future predictions.;A prototype single element nutrient injector was remotely controlled by a Macintosh computer after each decision was made in order to implement the decision.