請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/78118
標題: | 時間稽延神經網路應用於蜂群預測之研究 A Time Delay Neural Network on the Prediction of Honey Bee Colony Activity |
作者: | Wei-Sheng Chen 陳韋勝 |
指導教授: | 江昭皚(Joe-Air Jiang) |
關鍵字: | 蜂群出入巢行為,時間稽延神經網路,外部輸入自動回歸模型,蜂群行為監測系統, Honey bee flight behavior,time delay neural network,NARX model,bee counter, |
出版年 : | 2016 |
學位: | 碩士 |
摘要: | 蜜蜂是自然界中最重要的植物授粉者,目前人類的食物有三分之一來自於開花植物,其中約有百分之八十需要蜜蜂協助授粉,此外地球上還有許多植物需要蜜蜂扮演傳媒的角色,故蜜蜂對於人類的農作物生產,及地球生態系平衡的影響力,占有舉足輕重的地位。近年來,世界各地陸續發現大量蜜蜂不明消失的現象,使得蜂農損失慘重,根據相關研究人員研究發現,大量蜂群消失之原因為工蜂無法回巢而凋亡,此現象稱為蜂群崩潰失調症 。
本研究欲開發一蜂群行為預測模型,其主要目地為針對蜂群每日之活動力進行預測,本研究首先研製一蜂群行為即時監測系統,此系統目的為監測蜂群每日出入蜂箱之行為,系統亦能自動化記錄當下之環境溫濕度與蜂箱內溫濕度;後端分析部分,所提出之方法為使用外部輸入非線性自動回歸模型之時間稽延神經網路 ,並透過輸入環境溫度參數及蜂群入巢頻率參數訓練模型;本研究採用移動視窗的方式下,研究時間將規劃成五個移動視窗,移動時間的長度為四週。此外,於每個移動視窗內之資料各自分為樣本內配適資料(訓練資料)及樣本外預測資料(測試資料)。本研究針對每個視窗內樣本內資料分別配適出適當之稽延神經網路預測模型,並評估模型建立後對樣本外資料的預測績效。至於針對模型的預測績效,本文分別以預測誤差及均方誤差來做比較,且本研究所提出之預測模型其預測誤差約在15%,故模型之準確率約為85%。 本研究之貢獻在於開發一蜂群行為預測模型,此模型將可提供蜂農及研究人員精準且客觀的預測數據,透過預測數據將可在蜂群之蜂勢開始崩潰或凋亡前進行防範及補救措施。 Honey bees play an important role in pollinating flowering plants. According to the survey, one third of the world’s food supply originates from flowering plants, about 80% require the assistance from bees for pollination. Thus honeybees have played an important role in crop production and have had a crucial influence in the world's ecological balance. However, in recent years, many countries around the world have witnessed a mysterious phenomenon that honeybee populations disappear one after another, causing a serious loss to beekeepers. According to related research, the majority of bee colonies have collapsed due to worker bees withering to death after being unable to locate their nests. It is a particular case of collapse of honey bee colonies and still unresolved. The purpose of this study is to develop a prediction model for honeybee flight behavior by using time-delay neural network (TDNN) forecasting. To achieve this goal, this study first develop the real-time honeybee behavior monitoring system. In order to quantify the number and the frequency of out-going and in-coming bees. This research develops a time-delay neural network with NARX (nonlinear autoregressive network with exogenous inputs) dynamic neural architecture. The effectiveness, feasibility and robustness of the proposed method are demonstrated on a real data set, which is the historical honey bee in-coming frequency data and ambient temperature data measured from bee counter. The proposed method has successfully achieved reasonable detection and prediction of non-linear interaction patterns of daily honey bee flight activity among process variables. The percentage error of the prediction result is less than 15%. That is the accurate of this neural network model is reached up to 85%. The prediction result helps beekeeper and researchers to have a better understanding of forager flight behavior and takes actions before colony collapse. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/78118 |
DOI: | 10.6342/NTU201602511 |
全文授權: | 有償授權 |
電子全文公開日期: | 2026-12-31 |
顯示於系所單位: | 生物機電工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-105-R03631035-1.pdf 目前未授權公開取用 | 2.86 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。