對電力業者來說,用戶之用電行為關係到用電負載之變化,也影響到供電成本的問題。以往對於用電行為之推估大多是屬於區域性的預測,並無針對單一用戶進行觀察。本研究主要針對多用戶之用電行為做叢集化之動作,期能以資料挖掘的方式比對單一用電戶間之用電行為,找出其間相似者,並將之分入同一叢集中。並在這之中找出適用於此種大量資料之挖掘動作的方法,使整個分析的動作可以用一種更快,而且亦足夠精確的方式進行。 在叢集化完成之後,為了能夠在有新用電戶加入時,能夠依其資料,將之分入適當的叢集中,本研究於實驗中也提到決策樹分類之做法,期能在尚未有用戶之用電行為發生時就能將之分入適當叢集。
Power usage is most concerned by power suppliers. Since it will affect the cost of power supply and peak load. Most of past research focused on forecasting peak load of a single area but didn’t care about single customer’s power usage. Our research focuses on multiple-customer-clustering. We try to compare every single customer’s power usage and find out most similar customers and assign them to same cluster. We also try to found a method that can make the clustering faster and still accurate. After clustering, we want to classification new customer into a cluster fit to them. We discuss the ability of using decision-tree to classify a new customer into a suitable cluster by using their base data.