現今,甘藷價值越來越高,不僅是最佳養身代表之一,還是重要糧食與能源作物。但是要如何在有限的種植土地中,有效提升甘藷的產量與品質,甚至是預估還未收成的量與質,這對企業經營管理是很重要的。本研究藉由台灣著名甘藷生產供應商瓜瓜園企業股份有限公司所提供的種植履歷樣本資料與某地區農改場提供的農業氣象資料做結合,再加上土地相關資料,利用五種機器學習方法在甘藷種植的各個期間建立預測模型以進行分析。我們發現在五種方法中,隨機森林在預測甘藷良率準確度上,隨著種植由前到後期準確率穩定提升,同時在最後一期的預測準確度上也表現最佳。我們的預測模型也能找出在種植期間的重要輸入變數。另外,我們在預測產量的準確度上,尚未找到如預測良率相似的結果。我們相信未來如能取得更多年度的生產資料,不論預測良率或是預測產量都會得到有價值的預測模型,能夠提供農企業管理者一個重要的營運管理工具,使傳統農業能進步到精準農業,讓甘藷種植更有效率。
Nowadays, the value of sweet potatoes gets higher and higher. It becomes not only one of the best representatives of health preservation, but also a significant food and energy crop. However, efficiently increasing its output and quality, even making estimation on the output and quality of the part which has not been harvested, are important for the operation and management of enterprises. The research combines the sample date of planting experience provided by the K.K.Orchard of Agriculture Production and Marketing Groups with the agro-meteorological data provided by an agriculture improving field. And along with data related to lands, it establishes prediction model on every period of sweet potato planting for analysis based on five kinds of machine learning methods. It finds out that in the five methods, the degree of accuracy on estimating the yield rate of sweet potatoes in random forest is increasing from the earlier planting stage to the later stage. Meanwhile the best degree of accuracy is obtained at the last stage. The prediction model can also figure out the important input variables of the planting. Besides, it has not achieved a result similar to the estimated yield in the aspect of the degree of accuracy estimating the output. If more production data can be obtained in the future, there will be valuable prediction model no matter in estimating yield or output that can provide a significant operation management tool for farmers and agriculture enterprise managers, thus facilitating the traditional agriculture to develop into a precision agriculture to make higher efficiency of sweet potato planting.