本研究試以雙向長短期記憶模型(Bidirectional Long Short-Term memory)加上注意力機制(Attention)嘗試解決時間序列資料,跨越多時間區間上的障礙。並嘗試創造出一模型架構預測作物產期,來解決傳統的耕作上,對於產期預測方面多以耕作者的經驗來預測之問題,因作物生長需考慮天氣環境等特徵對農作物生長方面造成的影響,所以本研究採用中央氣象局所提供之天氣資料結合政府產銷履歷資料庫,以青江菜為例,結合人工智慧及機器學習方法,並應用在傳統農業上,解決了傳統農業無法以資訊化的方式來預測的問題,且也結合Attention方法改善了傳統RNNs模型問題。
This study attempts to solve the span problem of time series data with Bidirectional Long Short-Term Memory and Attention funtion. A model structure is create to predict the crop production period. At present the traditional farming’s production period mostly based on the experience of farmer, so the study try to solve this problem. In addition the crop growth needs to consider the characteristics of the weather environment. Therefore this study uses the weather data provided by the Central Meteorological Administration and government’s crop period database. The study will use Qingjiang cuisine as an example, combined with artificial intelligence and machine learning methods. And solves the problem of traditional agriculture also improves the RNNs model problem by combining attention methods. The study applied to traditional agriculture makes that more evolved and intelligent in the future.