本研究應用倒傳遞類神經網路(Back-Propagation Neural Network, BNN)與多元線性迴歸(Multiple Linear Regression, MLR)建立臺中縣營建工地空氣品質預測模式,使用之變數為懸浮微粒(PM2.5)、懸浮微粒(PM10)、建築(房屋)工程(SRC)、建築(房屋)工程(RC)、隧道工程及其他營建工程,以2008年1月至11月數據建立最適化網路後,對2008年12月之PM2.5作預測,並比較BNN與MLR之預測結果。 本研究利用BNN及MLR模擬預測臺中縣PM2.5之研究結果顯示,BNN及MLR模型對於PM2.5之濃度及變動趨勢皆可掌握,其中MLR於大里之測試及預測效果較BNN為佳,預測之相關係數以大里(6V1)及沙鹿(6V1、5V1、4V1)之0.47為最高,以豐原(4V1、3 V1、2 V1)之0.29為最低; BNN於豐原及沙鹿之測試及預測效果較MLR為佳,預測之相關係數以豐原(6V1、4V1~2V1)之0.94為最高,以大里(4V1、3V1)之0.40為最低。BNN訓練之相關係數介於0.73~0.95,MLR測試之相關係數介於0.68~0.76,模擬預測結果BNN之相關係數介於0.4~0.94,MLR之相關係數介於0.29~0.47,整体而言,BNN預測結果優於MLR。
This research uses Back-Propagation Neural Network(BNN)and Multiple Linear Regression(MLR)to establish construction sites’ air quality forecasting module in Taichung County. The variables are PM2.5, PM10, SRC, RC, tunnel constructions and other construction works. By using the optimized network established from data of January to November 2008, a forecast was produced using BNN and MLR for the result in 2008. By using BNN and MLR, this research have produced a simulated forecast for PM2.5 in Taichung County. Both BNN and MLR forecasting models have showed capabilities in capturing the changes and trend of the PM2.5 concentration level. At DaLi MLR outperformed BNN in forecasting results in DaLi, the forecasted relative factor is highest in DaLi(6V1)and ShaLu(6V1, 5V1, 4V1)at 0.47 and lowest at FongYuang(4V1, 3V1, 2V1)at 0.29. BNN outperformed MLR in forecasting results in FongYuan and ShaLu, the forecasted relative factor is highest in FongYuan(4V1, 4V1~2V1)at 0.94 and lowest in DaLi(4V1, 3V1)at DaLi. BNN’s relative training factor is between 0.73~0.95 whereas MLR’s relative testing factor is between 0.68~0.76. The relative factor of simulated forecast is between 0.4~0.94 for BNN and 0.29~0.47 for MLR, overall speaking BNN’s forecast is better than MLR.