Data selection and fuzzy-rules generation for short-term load forecasting using ANFIS

Abstract

This paper focused on data analysis, with aim of determining the actual variables that affect the load consumption in short term electric load forecasting. Correlation analysis was used to determine how the load consumption is related to the forecasting variables (model inputs), and hypothesis test was used to justify the correlation coefficient of each variable. Three different models based on data selection criteria where tested using Adaptive Neuro-Fuzzy Inference System (ANFIS). Subtractive Clustering (SC) and Fuzzy c-means (FCM) rules generation algorithms ware compared in all the three models. It was observed that forecasting using Hypothesis test data with SC algorithm gave better accuracy compared to the other two approaches. But FCM algorithm is faster in all the three approaches. In conclusion, hypothesis test on the correlation coefficient of the data is a commendable practice for data selection and analysis in short-term load forecasting.