應用開放資料預測農產品菜價之研究：以甘藍為例

Abstract

農產品價格的波動影響著我們日常生活。近年來劇烈的氣候變化更加劇菜價漲跌。本研究目的為農產品交易資訊透明化做出貢獻，研究流程分成三大部分：自動化資料擷取與特徵工程、資料視覺化、農產品價格預測模型。本研究首先搜集農產品交易開放資料、氣象開放資料和颱風警報資料，並透過資料清洗及各項特徵工程方法來整理資料和特徵建構，經過多次試算與調整後，再將資料擷取、特徵工程等步驟整合並撰寫成自動化擷取程式，使得未來更新資料能夠不依靠人工便可自動化更新；此外本論文提出農產品交易訊息視覺化方法，藉由視覺化圖形彼此交互比對，使得大眾能夠直觀地觀察、分析龐大且繁雜的數據，最後使用類神經網路之LSTM（長短期記憶模型）設計價格預測模型，預測全國最大宗蔬菜-甘藍價格。

The purpose of this study is to predict vegetable prices in the wholesale markets using governmental open data. We collected various governmental open data in Taiwan, including agricultural product prices in wholesale markets and climate statistics of each county in Taiwan. In order to develop our prediction model, we organize and construct vegetable price data features using data cleansing and feature engineering. We also develop automated data extraction programs to constantly download the new data without human intervention. In addition, a visualization scheme is developed for these complicated agricultural transaction data. Our prediction model, LSTM (Long short-term memory) architecture, is developed using Deep Learning RNN (Recurrent Neural Network) architecture. A preliminary vegetable price prediction experiment is conducted.