本研究以D公司位於桃園市楊梅區之大型自動化恆溫倉儲（AS/RS）為例，以連續7日之溫度測繪（Temperature mapping）紀錄，建立長短期記憶網路（Long Short Term Memory Network, LSTM）及循環神經網路（Recurrent Neural Network，RNN）模型。研究利用前150小時AS/RS中代表點位EDLM（Electronic data logging monitor，電子式溫度記錄器）的溫度數據，以LSTM模型預測未來18小時代表點位之溫度；而後再利用RNN模型將LSTM的預測結果，估計同空間其他多數點位未來18小時溫度趨勢。為了驗證其預測結果，將建築物管理系統（Building Management System，簡稱BMS）中擷取Real-time Sensor所紀錄之溫度資料來做驗證，並比較LSTM與RNN的最大絕對誤差（Max AE）。 結果顯示，無論模型在Max AE 還是RMSE上，上述模型與方法皆能符合RMSE小於0.20，且Max AE小於儀器MPE（Maximum Permissible Error，最大允許誤差）0.50℃。結果間接地證實，未來可以少數的Real-time Sensor監測，取代原先醫療藥品倉中需使用EDLM大量佈點來執行定期溫度測繪評估。最後依據本次個案分析提出觀察及未來模型可考量的變因，做為後續研究方向參考。;In this paper, we use D company’s large-scale Automated storage system (Automated Storage / Retrieval System, AS/RS) warehouse in Taoyuan City Yangmei District as example, propose a deep learning method for temperature estimation. The method is based on Long Short-Term Memory (LSTM) model and Recurrent Neural Network (RNN) by using 7-days temperature mapping records. Using LSTM model with prior 150 hours key spots’ EDLM（Electronic data logging monitor）temperature data to predict future 18 hours temperature, then using RNN model with LSTM key spots’ predict data to estimate future 18 hours temperature of multiple spots in AS/RS. To validate the estimation result, we use the Real-time sensor temperature record collected in Building Management System (BMS) as validation data. Comparing the Max AE (Max Absolute Error) of LSTM and RNN model. According to the result, no matter checking Models’ Max AE or RMSE, the methods all comply RMSE less than 0.20 and Max AE less than MPE (Maximum Permissible Error) which is 0.50℃. That also proves the feasibility of using real-time sensors as temperature mapping method, to replace large quantity of EDLM in Pharmaceutical warehouse. At the end of this paper, we describe some observations and potential variation, as the reference information for future studies.
