基於深度學習的物聯網設備容量限制服務業之碳足跡排放預測 - 以自助洗衣業為例

Abstract

面對全球減碳壓力與淨零排放目標，碳足跡管理及預測議題漸趨重要；雖然製造業等領域在過去存在相關研究，但對於如餐廳、自助洗衣店等容量限制服務業，相關研究仍相對稀少，使產業難以量化自身碳足跡排放量，也使政策制定缺乏實證依據。

本研究以自助洗衣業為範例，提出一個碳足跡計算及管理與預測的框架。透過業者提供的IoT 資料，整合機台的電力、水與廢水三項主要耗用量，並提出以排放係數為基礎的碳足跡計算方式；其次，再以歷史營運紀錄訓練時間序列模型，預測未來十二個月碳足跡的趨勢；考量新店面或新機台常面臨資料不足，本研究也設計分群模型，依設備使用型態與環境條件尋找相似群組，而分開處理不同量級的資源，則可以進一步提供模型穩定度及精確度。

研究結果顯示，所建構的碳足跡時間序列模型與分群模型均具良好準確度，該框架不僅可協助自助洗衣業據此制定減排策略，也能進一步延伸至餐廳、電動車充電椿等其他容量限制服務領域，做為永續營運之參考。

Facing growing global pressure to reduce carbon emissions and achieve net-zero targets, carbon-footprint management and forecasting have become increasingly important. Although previous studies exist in sectors such as manufacturing, research on capacitated service industries—such as restaurants and self-service laundromats—remains relatively scarce, making it difficult for these industries to quantify their own carbon-footprint emissions and leaving policymakers without empirical evidence.

Using the self-service laundry industry as an example, this study proposes a framework for carbon-footprint calculation, management, and forecasting. IoT data supplied by operators are used to integrate the three main resource consumptions of each machine—electricity, water, and wastewater—and an emission-factor-based method for calculating the carbon footprint is presented. Historical operational records then train a time-series model to predict carbon-footprint trends for the next twelve months. Because new stores or newly installed machines often lack sufficient data, a clustering model is designed to identify similar groups based on equipment usage patterns and environmental conditions; handling resources of different scales separately further enhances model stability and accuracy.

The results show that both the time-series model and the clustering model achieve good accuracy. The framework can not only help the self-service laundry industry develop emission-reduction strategies but can also be extended to other capacitated service sectors—such as restaurants and EV charging stations—as a reference for sustainable operations.