Substation mid-term electric load forecasting by knowledge-based method

Abstract

In this paper, a novel knowledge-based method is proposed for mid-term load forecasting (MTLF) in electric substations. Since knowledge-based methods rely on the similar day strategy for estimating the load profile, novel techniques are proposed to properly select these days from historical datasets. As the maintenance of substations is inevitable during some days of the year, it can affect the load shapes and make them abnormal. Therefore, some errors occur in MTLF if similar days are selected from the days with abnormal load shapes. To tackle this problem in MTLF, a pre-processing procedure based on the flag concept is conducted to separate the misleading data causing the estimation error. In addition, a new categorization of historical data is proposed in order to select days with more similar load shapes. In this procedure, the effects of neighbor substation maintenance, holidays, and special cultural days such as Ramadan are considered in order to select similar days more carefully. Finally, the hourly load profile is forecasted using a linear equation. The performance of the proposed MTLF method is evaluated in a target substation in Tehran Regional Electric Company in Iran. The results show that each of the neglecting substation maintenance days with abnormal load shapes from the similar day selection process and monthly–weekly window of data can reduce the error of forecasting by about 8%. In addition, neglecting similar days, on which the connected neighbor substations have maintenance, can reduce 5% of forecasting error. It means that these considerations are very important and can be used for improving the results of load forecasting approaches. To prove the advantage of the proposed method over other load forecasting methods, the results are compared with those of the fuzzy-based load forecasting method, which is a well-known approach. The forecasting results are improved by approximately 4% compared with those of the recent approach using the fuzzy method, which shows the applicability and superiority of the proposed method.