以類電磁為基礎之類神經綱路技術應用於太陽能預測

Abstract

本論文探討使用類電磁演算法優化類神經網路並應用於日射量預測問題。首先，建置一日射量資料平台以收集日射量預測所需相關數據，該日射量資料平台設備包括日射計、單板電腦(single-board computer)、攝影機與雲端伺服器等。藉由整合該日射量資料平台設備以即時擷取日射量與當前天空之影像圖，並儲存於雲端伺服器MySQL資料庫。利用該資料庫日射量與天空之影像圖資料，建立一即時日射量類神經網路預測系統，該日射量類神經網路預測系統可領前1至6小時預測日射量。為了使預測結果更加準確，透過類電磁演算法的改造與改良，使該演算法可以用於優化該類神經網路預測系統。最後，將本文所提之類電磁神經網路和傳統類神經網路進行預測誤差比較，以驗證本文所提之類電磁神經網路之效能。

In the thesis, a solar irradiance forecasting system is developed by using an electromagnetism-like neural network. The hardware of the solar irradiance forecasting system includes a pyranometer, single-board computer, webcam and cloud server. The hardware devices are used to collect and store the solar data, including solar irradiance data and sky images. The electromagnetism-like mechanism algorithm is improved and involved in the neural network in order to increase the forecasting efficiency. The input features of the neural network includes real-time sky and solar irradiance data and past solar irradiance data. Based on the trained neural network, the solar irradiance forecasting system can calculate the hourly 1-6 hours ahead solar irradiance values. Finally, some comparison results are given to verify the efficiency of solar irradiance forecasting system.