長崎大学経済学部

研究活動

No.2004-05 (December 2004)
An Approach of Electric Power Demand Forecasting using Data-Mining Method:
a case study of application of Data-Mining technique to improvement of decision-making
Toshio SUGIHARA (Professor, Faculty of Economics, Nagasaki University)

Abstract:
  In this paper, for the aim to build the management plan of stable electric power supply and accommodation, the monthly electric demand prediction approach with dynamic and adaptive mechanism connected to the business environment is proposed.
  The proposed prediction adopts the Kalman-Filter as the basic prediction scheme and possess two characteristics stated below. One is the state-space built with the principal component time-series integrated by time-series PCA (Principal Component Method) from multi business indices related the targeted time-series. The other is the self-organized auto-updating of the state-space by structured neural network.
  The proposed scheme shows considerably accurate prediction than any other models with single variable time-series and the obvious effect appears to the high accuracy by adopting time-series PCA as a Data-Mining technique. From these effects, the proposed prediction scheme might be considered to give an improvement to the stable electric power supply and accommodation. And this prediction scheme can be applied to various management areas, so it might be considered to be an effective method for decision-making support.

Keywords:
Electric power supply; Knowledge extraction; Data-Mining technique;
Kalman-Filter processing; Self-organized state-space;
Principal component time-series; Batch-Sequential method