- 著者
-
蔵田 憲次
- 出版者
- The Society of Agricultural Meteorology of Japan
- 雑誌
- 農業気象 (ISSN:00218588)
- 巻号頁・発行日
- vol.44, no.3, pp.181-186, 1988-12-10 (Released:2010-02-25)
- 参考文献数
- 6
- 被引用文献数
-
1
3
Computers have been applied to crop management in protected cultivation and have proved to be of great use. However, in the present computer systems, control algorithm is fixed and users can only change the setpoints of the controlled variables. Crop management is strongly influenced by the local conditions of climate, soil, etc. and well trained and eager growers manage their crops in the way which is suitable under the given local conditions and is obtained through the long period experience.This study aims at developing a system which learns grower's managing methods (rules, hereafter) through measuring environmental factors, crop status, if possible, and grower's behavior.After learning the grower's rules, the learned rules will be applied to the automatic management.Thus, the grower will be released from the management labor without losing their own personal preference and principles in crop management. This system can also be applied to the rule aquisition and analysis of well trained growers. This report presents considerations on the characteristics of the growers' rules and the possibilities and problems of the simple learning algorism developed (K-algorithm, hereafter), which will be the framework of the system we aim at. Details of K-algorithm are given in another report (Kurata, 1988).Considerations on the grower's rules revealed that there are many difficult aspects for the computer in learning, for examples, the problem of fuzziness associated with the human feelings, the problem of noise and how to find out a function which the grower's rule has but not directly measurable. All these points are omitted in the present study and left for the future study.A simple simulation revealed that K-algorithm is very powerful in imitating the grower's behavior but the learned rules are somewhat longer than the assumed grower's rules, examples of which are listed Table 1. This redundant expression of the learned rule is due to some correlations among environmental factors, crops status and time. If we assume stochastic stationary state of these functions and their correlations, this brings about no problem in applying the learned rules to the automatic management.