用電量之預測－以用電兼具隨機性與週期性之設施為例

Prediction of electricity charge – An example of facilities involving both random and periodical varying

經常門經費之預估乃是物業管理重要工作，尤其是為了預算編列工作。傳統上經常門經費之預估一般係參照既有記錄，憑經驗預估或假設趨勢函數回歸預估，此種作法不但缺乏實質依據，也會因影響經常門經費的因素過於隨機消長且複雜，以致預估的誤差無法掌控。本文以傅立葉級數為理論依據，剖析具週期性特色的用電進而預測其未來電費，以達到準確預估經常門經費之目的。本文利用傅立葉級數分析既有用電量記錄，計算出組合該用電量記錄各正弦函數之頻率、振幅與相角差，挑選振幅較大之各正弦函數，捨棄其餘振幅較小之各正弦函數，以組合回用電量記錄之計算值，並預測未來之用電量，再以實際用電量驗證本分析技術之精確性。此技術可用來分析及預估各種具備週期性數據，如旅社住房率、設施維修費、餐廳消費人次、臨時工需求及停車位需求等。本研究需要既有數據，特點是量化數據，無人為主觀判定，適合短期（不適於長期）週期性數據之分析及預估。

Prediction for maintaining expenses is an important issue of property management, especially for budget planning. Normally, record in the pass is an important reference. By experience and regression based on assumed function, prediction is made, but lacks of solid basis. As the factors effecting the maintaining expenses vary randomly the prediction becomes complex, and the precision is out of control. This study bases theoretically on Fourier series to analyze the roughly periodical electricity expenses for predicting electricity charge in the future. The algorithm can be used to predict other kinds of maintaining expenses precisely. In this study, data of electricity charge of a building are analyzed by Fourier series to derived frequencies, amplitudes and angle differences of series of sine functions to form the data. Major sine functions are chosen and minor ones are abandoned to reconstruct the data as the prediction for the future. Prediction is verified experimentally to confirm the precision of the algorithm. This algorithm is capable of analyzing and predicting data involving periodical behavior or condition like lodging rate of a hotel, maintaining expense of a system, visiting man-time of a restaurant, requirement of short-term worker, space of a parking lot, etc. Based on recorded data, the quantified prediction involves no subjective judgment and is suitable for short-term prediction in the future, but not for long-term prediction.