A Statistical Data Selection Approach for Short-Term Load Forecasting Using Optimized ANFIS

Abstract

Volume of the forecasting data and good data analysis are the key factors that influence the accuracy of forecasting algorithm because it depends on data identification and model parameters. This paper focuses on data selection approach for short-term load forecasting. It involves formulating data selection algorithm to identify factors (variables) that influence energy demand at utility level. Correlation Analysis (CA) and Hypothesis Test (HT) are used in the selection, where Wavelet Transform (WT) is applied to bridge the gap between the forecasting variables. This results to three groups of data; data without CA, HT and WT, data with CA, HT but without WT and data with CA, HT and WT. An optimized adaptive neuro-fuzzy inference system (ANFIS) using Cuckoo Search Algorithm (CS) is used to conduct the forecasting. The essence is to reduce the computational difficulty associated with the gradient descent (GD) algorithm in traditional ANFIS. With the three data groups, it is observed that CHW data can give satisfactory results more than the NCNHNW and NCNHW data. Also the numerical results shows that CHW data selection approach can give a MAPE of 0.63 against the bench-mark approach with MAPE of 3.55. This indicates that it is good practice to select the actual data and process it before the forecasting.