Short-Term Load Forecasting Study Based On Data Mining

Short-term load forecasting is one of the most important routine works for power dispatch departments. The accuracy of load forecasting will exert direct effects on the safety, economy and stabilization of the power system running. There are a handful of factors which could affect power load that share the characteristics of non-linearity, complexity and hysteretic This paper employs data mining technique to mine the effects of variant factors on the load, forwards a new short-term-load algorithm which combines decision-tree and time series approaches, and achieves a short-term load forecasting system with DotNet architecture.Data-mining technique can extract potential knowledge and information that will be of interest from an extremely abundance of data. The paper summarizes theories and practices concerning data mining at the beginning. According to actual conditions of data forecasting, the paper particularizes two basic tasks in data mining, namely, exploring data analysis and forecasting modeling for classification, and employs them in load forecasting practically. Then a special short-term load-forecasting arithmetic is forwarded based on the former work, which can efficiently take into account weather effects on the load, and hence improve the accuracy of load forecasting.Portrait and transverse comparability are employed to distinguish and correct bad load data, while wavelet analysis and multiple-time-period analysis used to eliminate long-term increasing weights, thus reducing the impact of the high-speed load increase on the accuracy of load forecasting.With classification and regression tree algorithm, the paper analyzes the effects of non-load factors like weather on the load, achieves the knowledge expression of decision tree about the relation between the load and concerned factors as weather, and goes further to put forward a new short-term load-forecasting arithmetic which combines decision tree and time series approaches. With the new arithmetic, the mined knowledge can be efficiently used for short-term load forecasting, with the merits of high computing rate and relatively high accuracy. By doing so, not only related factors like weather are effectively involved, but the characteristics of time series are properly looked after. Based on the former work, a system of short-term load forecasting is realized with DotNet architecture. After elaborating on the need for short-term load forecasting and structure and function realization under the current conditions of electricity market, the paper analyzes and discusses some concrete problems in the system development process.In the end, with the analysis of the actual system-running data, the paper summarizes main factors which may lay impact on the load accuracy.