Automation of crop management by machine learning

在设施栽培中，要获得更高的产量，特别是在最佳控制环境因素或最佳管理作物方面需要做很多工作。该项目旨在开发一种系统，减少管理作物所需的工作量。作物的管理取决于当地条件，如地下水位或当地气候。因此，我们是否能将一个通用算法应用于这样一个相当局部规定的管理是值得怀疑的。考虑到这一点，我们正在开发一个系统，首先试图从管理前的测量条件（例如，管理）中学习农民的管理方法或原则（规则，以下简称规则）。气候条件）和农民的经营行为。系统从测量数据中归纳出规则，并将规则应用于管理中，使农民从管理劳动中解放出来。农民对环境因素的控制不一定基于实际测量值，但在大多数情况下，根据他们的感觉。农民也不像机器那样精确地行动。这意味着学习算法应该能够科普模糊和噪声，这将使学习策略变得非常困难。尽管作了相当大的努力，我们在这一点上还没有成功。然而，我们已经开发的学习算法被证明能够完全复制管理行为，如果它不包含模糊或噪声。该算法适用于大多数作物管理的情况。

In protected cultivation, much work is needed to get higher yield especially in controlling the environmental factors at the optima or in managing the crops optimally. This project aims at developping a system which reduced amount of work needed in managing the crops. Management of the crops are dependent on local conditions such as water table level or local climate. It is, therefore, questionable if we can apply a general algorithm to such a quite locally prescribed management. Taking this point into consideration we are developping a system which first tries to learn the methods or principles of management of the farmer (rules, hereafter) from the measured conditions prescrbing the management (eg. climate conditions) and the management behavior of the farmer. After inducing the rules from the measurements, the system applies the rules to the management, which releases the farmer from the management labor. Farmers control the environmental factors not necessarily based on the physically measured values but in most cases according to what they feel. Farmers also do not act as accurately as machines. These mean that the learning algoritjm should be able to cope with fuzziness and noise, which will make the learning strategy quite difficult. In spite of considerable effort we have not yet been successful in this point. However, the learning algorithm we have developped proved to be able to completely copy the management behavior, if it containes no fuzziness nor noise. This algorithm can be applied to most cases of the crop management.